4 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
6 * This program is free software; you can redistribute it and/or modify
7 * it under the terms of the GNU General Public License version 2 only,
8 * as published by the Free Software Foundation.
10 * This program is distributed in the hope that it will be useful, but
11 * WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
13 * General Public License version 2 for more details (a copy is included
14 * in the LICENSE file that accompanied this code).
16 * You should have received a copy of the GNU General Public License
17 * version 2 along with this program; If not, see
18 * http://www.gnu.org/licenses/gpl-2.0.html
23 * Copyright (c) 2003, 2010, Oracle and/or its affiliates. All rights reserved.
24 * Use is subject to license terms.
26 * Copyright (c) 2011, 2017, Intel Corporation.
29 * This file is part of Lustre, http://www.lustre.org/
30 * Lustre is a trademark of Sun Microsystems, Inc.
32 * lustre/llite/llite_lib.c
34 * Lustre Light Super operations
37 #define DEBUG_SUBSYSTEM S_LLITE
39 #include <linux/module.h>
40 #include <linux/random.h>
41 #include <linux/statfs.h>
42 #include <linux/time.h>
43 #include <linux/types.h>
44 #include <libcfs/linux/linux-uuid.h>
45 #include <linux/version.h>
47 #include <linux/user_namespace.h>
48 #ifdef HAVE_UIDGID_HEADER
49 # include <linux/uidgid.h>
51 #include <linux/security.h>
53 #include <uapi/linux/lustre/lustre_ioctl.h>
54 #include <lustre_ha.h>
55 #include <lustre_dlm.h>
56 #include <lprocfs_status.h>
57 #include <lustre_disk.h>
58 #include <uapi/linux/lustre/lustre_param.h>
59 #include <lustre_log.h>
60 #include <cl_object.h>
61 #include <obd_cksum.h>
62 #include "llite_internal.h"
64 struct kmem_cache *ll_file_data_slab;
67 #define log2(n) ffz(~(n))
70 static struct ll_sb_info *ll_init_sbi(void)
72 struct ll_sb_info *sbi = NULL;
74 unsigned long lru_page_max;
83 spin_lock_init(&sbi->ll_lock);
84 mutex_init(&sbi->ll_lco.lco_lock);
85 spin_lock_init(&sbi->ll_pp_extent_lock);
86 spin_lock_init(&sbi->ll_process_lock);
87 sbi->ll_rw_stats_on = 0;
90 pages = si.totalram - si.totalhigh;
91 lru_page_max = pages / 2;
93 /* initialize ll_cache data */
94 sbi->ll_cache = cl_cache_init(lru_page_max);
95 if (sbi->ll_cache == NULL) {
96 OBD_FREE(sbi, sizeof(*sbi));
100 sbi->ll_ra_info.ra_max_pages_per_file = min(pages / 32,
101 SBI_DEFAULT_READAHEAD_MAX);
102 sbi->ll_ra_info.ra_max_pages = sbi->ll_ra_info.ra_max_pages_per_file;
103 sbi->ll_ra_info.ra_max_read_ahead_whole_pages = -1;
105 sbi->ll_flags |= LL_SBI_VERBOSE;
106 #ifdef ENABLE_CHECKSUM
107 sbi->ll_flags |= LL_SBI_CHECKSUM;
110 sbi->ll_flags |= LL_SBI_FLOCK;
113 #ifdef HAVE_LRU_RESIZE_SUPPORT
114 sbi->ll_flags |= LL_SBI_LRU_RESIZE;
116 sbi->ll_flags |= LL_SBI_LAZYSTATFS;
118 for (i = 0; i <= LL_PROCESS_HIST_MAX; i++) {
119 spin_lock_init(&sbi->ll_rw_extents_info.pp_extents[i].
121 spin_lock_init(&sbi->ll_rw_extents_info.pp_extents[i].
125 /* metadata statahead is enabled by default */
126 sbi->ll_sa_running_max = LL_SA_RUNNING_DEF;
127 sbi->ll_sa_max = LL_SA_RPC_DEF;
128 atomic_set(&sbi->ll_sa_total, 0);
129 atomic_set(&sbi->ll_sa_wrong, 0);
130 atomic_set(&sbi->ll_sa_running, 0);
131 atomic_set(&sbi->ll_agl_total, 0);
132 sbi->ll_flags |= LL_SBI_AGL_ENABLED;
133 sbi->ll_flags |= LL_SBI_FAST_READ;
134 sbi->ll_flags |= LL_SBI_TINY_WRITE;
137 sbi->ll_squash.rsi_uid = 0;
138 sbi->ll_squash.rsi_gid = 0;
139 INIT_LIST_HEAD(&sbi->ll_squash.rsi_nosquash_nids);
140 init_rwsem(&sbi->ll_squash.rsi_sem);
142 /* Per-filesystem file heat */
143 sbi->ll_heat_decay_weight = SBI_DEFAULT_HEAT_DECAY_WEIGHT;
144 sbi->ll_heat_period_second = SBI_DEFAULT_HEAT_PERIOD_SECOND;
148 static void ll_free_sbi(struct super_block *sb)
150 struct ll_sb_info *sbi = ll_s2sbi(sb);
154 if (!list_empty(&sbi->ll_squash.rsi_nosquash_nids))
155 cfs_free_nidlist(&sbi->ll_squash.rsi_nosquash_nids);
156 if (sbi->ll_cache != NULL) {
157 cl_cache_decref(sbi->ll_cache);
158 sbi->ll_cache = NULL;
160 OBD_FREE(sbi, sizeof(*sbi));
165 static int client_common_fill_super(struct super_block *sb, char *md, char *dt,
166 struct vfsmount *mnt)
168 struct inode *root = NULL;
169 struct ll_sb_info *sbi = ll_s2sbi(sb);
170 struct obd_statfs *osfs = NULL;
171 struct ptlrpc_request *request = NULL;
172 struct obd_connect_data *data = NULL;
173 struct obd_uuid *uuid;
174 struct md_op_data *op_data;
175 struct lustre_md lmd;
177 int size, err, checksum;
180 sbi->ll_md_obd = class_name2obd(md);
181 if (!sbi->ll_md_obd) {
182 CERROR("MD %s: not setup or attached\n", md);
196 /* pass client page size via ocd_grant_blkbits, the server should report
197 * back its backend blocksize for grant calculation purpose */
198 data->ocd_grant_blkbits = PAGE_SHIFT;
200 /* indicate MDT features supported by this client */
201 data->ocd_connect_flags = OBD_CONNECT_IBITS | OBD_CONNECT_NODEVOH |
202 OBD_CONNECT_ATTRFID | OBD_CONNECT_GRANT |
203 OBD_CONNECT_VERSION | OBD_CONNECT_BRW_SIZE |
204 OBD_CONNECT_SRVLOCK | OBD_CONNECT_TRUNCLOCK|
205 OBD_CONNECT_MDS_CAPA | OBD_CONNECT_OSS_CAPA |
206 OBD_CONNECT_CANCELSET | OBD_CONNECT_FID |
207 OBD_CONNECT_AT | OBD_CONNECT_LOV_V3 |
208 OBD_CONNECT_VBR | OBD_CONNECT_FULL20 |
209 OBD_CONNECT_64BITHASH |
210 OBD_CONNECT_EINPROGRESS |
211 OBD_CONNECT_JOBSTATS | OBD_CONNECT_LVB_TYPE |
212 OBD_CONNECT_LAYOUTLOCK | OBD_CONNECT_PINGLESS|
213 OBD_CONNECT_MAX_EASIZE |
214 OBD_CONNECT_FLOCK_DEAD |
215 OBD_CONNECT_DISP_STRIPE | OBD_CONNECT_LFSCK |
216 OBD_CONNECT_OPEN_BY_FID |
217 OBD_CONNECT_DIR_STRIPE |
218 OBD_CONNECT_BULK_MBITS | OBD_CONNECT_CKSUM |
219 OBD_CONNECT_SUBTREE |
220 OBD_CONNECT_MULTIMODRPCS |
221 OBD_CONNECT_GRANT_PARAM |
222 OBD_CONNECT_SHORTIO | OBD_CONNECT_FLAGS2;
224 data->ocd_connect_flags2 = OBD_CONNECT2_DIR_MIGRATE |
225 OBD_CONNECT2_SUM_STATFS |
227 OBD_CONNECT2_LOCK_CONVERT |
228 OBD_CONNECT2_ARCHIVE_ID_ARRAY |
231 #ifdef HAVE_LRU_RESIZE_SUPPORT
232 if (sbi->ll_flags & LL_SBI_LRU_RESIZE)
233 data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
235 #ifdef CONFIG_FS_POSIX_ACL
236 data->ocd_connect_flags |= OBD_CONNECT_ACL | OBD_CONNECT_UMASK |
237 OBD_CONNECT_LARGE_ACL;
240 data->ocd_cksum_types = obd_cksum_types_supported_client();
242 if (OBD_FAIL_CHECK(OBD_FAIL_MDC_LIGHTWEIGHT))
243 /* flag mdc connection as lightweight, only used for test
244 * purpose, use with care */
245 data->ocd_connect_flags |= OBD_CONNECT_LIGHTWEIGHT;
247 data->ocd_ibits_known = MDS_INODELOCK_FULL;
248 data->ocd_version = LUSTRE_VERSION_CODE;
250 if (sb->s_flags & MS_RDONLY)
251 data->ocd_connect_flags |= OBD_CONNECT_RDONLY;
252 if (sbi->ll_flags & LL_SBI_USER_XATTR)
253 data->ocd_connect_flags |= OBD_CONNECT_XATTR;
256 /* Setting this indicates we correctly support S_NOSEC (See kernel
257 * commit 9e1f1de02c2275d7172e18dc4e7c2065777611bf)
259 sb->s_flags |= MS_NOSEC;
262 if (sbi->ll_flags & LL_SBI_FLOCK)
263 sbi->ll_fop = &ll_file_operations_flock;
264 else if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
265 sbi->ll_fop = &ll_file_operations;
267 sbi->ll_fop = &ll_file_operations_noflock;
269 /* always ping even if server suppress_pings */
270 if (sbi->ll_flags & LL_SBI_ALWAYS_PING)
271 data->ocd_connect_flags &= ~OBD_CONNECT_PINGLESS;
273 obd_connect_set_secctx(data);
275 #if defined(CONFIG_SECURITY)
276 data->ocd_connect_flags2 |= OBD_CONNECT2_SELINUX_POLICY;
279 data->ocd_brw_size = MD_MAX_BRW_SIZE;
281 err = obd_connect(NULL, &sbi->ll_md_exp, sbi->ll_md_obd,
282 &sbi->ll_sb_uuid, data, sbi->ll_cache);
284 LCONSOLE_ERROR_MSG(0x14f, "An MDT (md %s) is performing "
285 "recovery, of which this client is not a "
286 "part. Please wait for recovery to complete,"
287 " abort, or time out.\n", md);
290 CERROR("cannot connect to %s: rc = %d\n", md, err);
294 sbi->ll_md_exp->exp_connect_data = *data;
296 err = obd_fid_init(sbi->ll_md_exp->exp_obd, sbi->ll_md_exp,
297 LUSTRE_SEQ_METADATA);
299 CERROR("%s: Can't init metadata layer FID infrastructure, "
300 "rc = %d\n", sbi->ll_md_exp->exp_obd->obd_name, err);
304 /* For mount, we only need fs info from MDT0, and also in DNE, it
305 * can make sure the client can be mounted as long as MDT0 is
307 err = obd_statfs(NULL, sbi->ll_md_exp, osfs,
308 ktime_get_seconds() - OBD_STATFS_CACHE_SECONDS,
309 OBD_STATFS_FOR_MDT0);
311 GOTO(out_md_fid, err);
313 /* This needs to be after statfs to ensure connect has finished.
314 * Note that "data" does NOT contain the valid connect reply.
315 * If connecting to a 1.8 server there will be no LMV device, so
316 * we can access the MDC export directly and exp_connect_flags will
317 * be non-zero, but if accessing an upgraded 2.1 server it will
318 * have the correct flags filled in.
319 * XXX: fill in the LMV exp_connect_flags from MDC(s). */
320 valid = exp_connect_flags(sbi->ll_md_exp) & CLIENT_CONNECT_MDT_REQD;
321 if (exp_connect_flags(sbi->ll_md_exp) != 0 &&
322 valid != CLIENT_CONNECT_MDT_REQD) {
325 OBD_ALLOC_WAIT(buf, PAGE_SIZE);
326 obd_connect_flags2str(buf, PAGE_SIZE,
327 valid ^ CLIENT_CONNECT_MDT_REQD, 0, ",");
328 LCONSOLE_ERROR_MSG(0x170, "Server %s does not support "
329 "feature(s) needed for correct operation "
330 "of this client (%s). Please upgrade "
331 "server or downgrade client.\n",
332 sbi->ll_md_exp->exp_obd->obd_name, buf);
333 OBD_FREE(buf, PAGE_SIZE);
334 GOTO(out_md_fid, err = -EPROTO);
337 size = sizeof(*data);
338 err = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_CONN_DATA),
339 KEY_CONN_DATA, &size, data);
341 CERROR("%s: Get connect data failed: rc = %d\n",
342 sbi->ll_md_exp->exp_obd->obd_name, err);
343 GOTO(out_md_fid, err);
346 LASSERT(osfs->os_bsize);
347 sb->s_blocksize = osfs->os_bsize;
348 sb->s_blocksize_bits = log2(osfs->os_bsize);
349 sb->s_magic = LL_SUPER_MAGIC;
350 sb->s_maxbytes = MAX_LFS_FILESIZE;
351 sbi->ll_namelen = osfs->os_namelen;
352 sbi->ll_mnt.mnt = current->fs->root.mnt;
354 if ((sbi->ll_flags & LL_SBI_USER_XATTR) &&
355 !(data->ocd_connect_flags & OBD_CONNECT_XATTR)) {
356 LCONSOLE_INFO("Disabling user_xattr feature because "
357 "it is not supported on the server\n");
358 sbi->ll_flags &= ~LL_SBI_USER_XATTR;
361 if (data->ocd_connect_flags & OBD_CONNECT_ACL) {
363 sb->s_flags |= MS_POSIXACL;
365 sbi->ll_flags |= LL_SBI_ACL;
367 LCONSOLE_INFO("client wants to enable acl, but mdt not!\n");
369 sb->s_flags &= ~MS_POSIXACL;
371 sbi->ll_flags &= ~LL_SBI_ACL;
374 if (data->ocd_connect_flags & OBD_CONNECT_64BITHASH)
375 sbi->ll_flags |= LL_SBI_64BIT_HASH;
377 if (data->ocd_connect_flags & OBD_CONNECT_LAYOUTLOCK)
378 sbi->ll_flags |= LL_SBI_LAYOUT_LOCK;
380 if (obd_connect_has_secctx(data))
381 sbi->ll_flags |= LL_SBI_FILE_SECCTX;
383 if (data->ocd_ibits_known & MDS_INODELOCK_XATTR) {
384 if (!(data->ocd_connect_flags & OBD_CONNECT_MAX_EASIZE)) {
385 LCONSOLE_INFO("%s: disabling xattr cache due to "
386 "unknown maximum xattr size.\n", dt);
387 } else if (!sbi->ll_xattr_cache_set) {
388 /* If xattr_cache is already set (no matter 0 or 1)
389 * during processing llog, it won't be enabled here. */
390 sbi->ll_flags |= LL_SBI_XATTR_CACHE;
391 sbi->ll_xattr_cache_enabled = 1;
395 sbi->ll_dt_obd = class_name2obd(dt);
396 if (!sbi->ll_dt_obd) {
397 CERROR("DT %s: not setup or attached\n", dt);
398 GOTO(out_md_fid, err = -ENODEV);
401 /* pass client page size via ocd_grant_blkbits, the server should report
402 * back its backend blocksize for grant calculation purpose */
403 data->ocd_grant_blkbits = PAGE_SHIFT;
405 /* indicate OST features supported by this client */
406 data->ocd_connect_flags = OBD_CONNECT_GRANT | OBD_CONNECT_VERSION |
407 OBD_CONNECT_REQPORTAL | OBD_CONNECT_BRW_SIZE |
408 OBD_CONNECT_CANCELSET | OBD_CONNECT_FID |
409 OBD_CONNECT_SRVLOCK | OBD_CONNECT_TRUNCLOCK|
410 OBD_CONNECT_AT | OBD_CONNECT_OSS_CAPA |
411 OBD_CONNECT_VBR | OBD_CONNECT_FULL20 |
412 OBD_CONNECT_64BITHASH | OBD_CONNECT_MAXBYTES |
413 OBD_CONNECT_EINPROGRESS |
414 OBD_CONNECT_JOBSTATS | OBD_CONNECT_LVB_TYPE |
415 OBD_CONNECT_LAYOUTLOCK |
416 OBD_CONNECT_PINGLESS | OBD_CONNECT_LFSCK |
417 OBD_CONNECT_BULK_MBITS | OBD_CONNECT_SHORTIO |
418 OBD_CONNECT_FLAGS2 | OBD_CONNECT_GRANT_SHRINK;
420 /* The client currently advertises support for OBD_CONNECT_LOCKAHEAD_OLD so it
421 * can interoperate with an older version of lockahead which was released prior
422 * to landing in master. This support will be dropped when 2.13 development
423 * starts. At the point, we should not just drop the connect flag (below), we
424 * should also remove the support in the code.
426 * Removing it means a few things:
427 * 1. Remove this section here
428 * 2. Remove CEF_NONBLOCK in ll_file_lockahead()
429 * 3. Remove function exp_connect_lockahead_old
430 * 4. Remove LDLM_FL_LOCKAHEAD_OLD_RESERVED in lustre_dlm_flags.h
432 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 12, 50, 0)
433 data->ocd_connect_flags |= OBD_CONNECT_LOCKAHEAD_OLD;
436 data->ocd_connect_flags2 = OBD_CONNECT2_LOCKAHEAD;
438 if (!OBD_FAIL_CHECK(OBD_FAIL_OSC_CONNECT_GRANT_PARAM))
439 data->ocd_connect_flags |= OBD_CONNECT_GRANT_PARAM;
441 /* OBD_CONNECT_CKSUM should always be set, even if checksums are
442 * disabled by default, because it can still be enabled on the
443 * fly via /sys. As a consequence, we still need to come to an
444 * agreement on the supported algorithms at connect time
446 data->ocd_connect_flags |= OBD_CONNECT_CKSUM;
448 if (OBD_FAIL_CHECK(OBD_FAIL_OSC_CKSUM_ADLER_ONLY))
449 data->ocd_cksum_types = OBD_CKSUM_ADLER;
451 data->ocd_cksum_types = obd_cksum_types_supported_client();
453 #ifdef HAVE_LRU_RESIZE_SUPPORT
454 data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
456 /* always ping even if server suppress_pings */
457 if (sbi->ll_flags & LL_SBI_ALWAYS_PING)
458 data->ocd_connect_flags &= ~OBD_CONNECT_PINGLESS;
460 CDEBUG(D_RPCTRACE, "ocd_connect_flags: %#llx ocd_version: %d "
461 "ocd_grant: %d\n", data->ocd_connect_flags,
462 data->ocd_version, data->ocd_grant);
464 sbi->ll_dt_obd->obd_upcall.onu_owner = &sbi->ll_lco;
465 sbi->ll_dt_obd->obd_upcall.onu_upcall = cl_ocd_update;
467 data->ocd_brw_size = DT_MAX_BRW_SIZE;
469 err = obd_connect(NULL, &sbi->ll_dt_exp, sbi->ll_dt_obd,
470 &sbi->ll_sb_uuid, data, sbi->ll_cache);
472 LCONSOLE_ERROR_MSG(0x150, "An OST (dt %s) is performing "
473 "recovery, of which this client is not a "
474 "part. Please wait for recovery to "
475 "complete, abort, or time out.\n", dt);
478 CERROR("%s: Cannot connect to %s: rc = %d\n",
479 sbi->ll_dt_exp->exp_obd->obd_name, dt, err);
483 sbi->ll_dt_exp->exp_connect_data = *data;
485 /* Don't change value if it was specified in the config log */
486 if (sbi->ll_ra_info.ra_max_read_ahead_whole_pages == -1)
487 sbi->ll_ra_info.ra_max_read_ahead_whole_pages =
488 max_t(unsigned long, SBI_DEFAULT_READAHEAD_WHOLE_MAX,
489 (data->ocd_brw_size >> PAGE_SHIFT));
491 err = obd_fid_init(sbi->ll_dt_exp->exp_obd, sbi->ll_dt_exp,
492 LUSTRE_SEQ_METADATA);
494 CERROR("%s: Can't init data layer FID infrastructure, "
495 "rc = %d\n", sbi->ll_dt_exp->exp_obd->obd_name, err);
499 mutex_lock(&sbi->ll_lco.lco_lock);
500 sbi->ll_lco.lco_flags = data->ocd_connect_flags;
501 sbi->ll_lco.lco_md_exp = sbi->ll_md_exp;
502 sbi->ll_lco.lco_dt_exp = sbi->ll_dt_exp;
503 mutex_unlock(&sbi->ll_lco.lco_lock);
505 fid_zero(&sbi->ll_root_fid);
506 err = md_get_root(sbi->ll_md_exp, get_mount_fileset(sb),
509 CERROR("cannot mds_connect: rc = %d\n", err);
510 GOTO(out_lock_cn_cb, err);
512 if (!fid_is_sane(&sbi->ll_root_fid)) {
513 CERROR("%s: Invalid root fid "DFID" during mount\n",
514 sbi->ll_md_exp->exp_obd->obd_name,
515 PFID(&sbi->ll_root_fid));
516 GOTO(out_lock_cn_cb, err = -EINVAL);
518 CDEBUG(D_SUPER, "rootfid "DFID"\n", PFID(&sbi->ll_root_fid));
520 sb->s_op = &lustre_super_operations;
521 #ifdef HAVE_XATTR_HANDLER_FLAGS
522 sb->s_xattr = ll_xattr_handlers;
524 #if THREAD_SIZE >= 8192 /*b=17630*/
525 sb->s_export_op = &lustre_export_operations;
529 * XXX: move this to after cbd setup? */
530 valid = OBD_MD_FLGETATTR | OBD_MD_FLBLOCKS | OBD_MD_FLMODEASIZE;
531 if (sbi->ll_flags & LL_SBI_ACL)
532 valid |= OBD_MD_FLACL;
534 OBD_ALLOC_PTR(op_data);
536 GOTO(out_lock_cn_cb, err = -ENOMEM);
538 op_data->op_fid1 = sbi->ll_root_fid;
539 op_data->op_mode = 0;
540 op_data->op_valid = valid;
542 err = md_getattr(sbi->ll_md_exp, op_data, &request);
544 OBD_FREE_PTR(op_data);
546 CERROR("%s: md_getattr failed for root: rc = %d\n",
547 sbi->ll_md_exp->exp_obd->obd_name, err);
548 GOTO(out_lock_cn_cb, err);
551 err = md_get_lustre_md(sbi->ll_md_exp, request, sbi->ll_dt_exp,
552 sbi->ll_md_exp, &lmd);
554 CERROR("failed to understand root inode md: rc = %d\n", err);
555 ptlrpc_req_finished(request);
556 GOTO(out_lock_cn_cb, err);
559 LASSERT(fid_is_sane(&sbi->ll_root_fid));
560 root = ll_iget(sb, cl_fid_build_ino(&sbi->ll_root_fid,
561 sbi->ll_flags & LL_SBI_32BIT_API),
563 md_free_lustre_md(sbi->ll_md_exp, &lmd);
564 ptlrpc_req_finished(request);
567 #ifdef CONFIG_FS_POSIX_ACL
569 posix_acl_release(lmd.posix_acl);
570 lmd.posix_acl = NULL;
573 err = IS_ERR(root) ? PTR_ERR(root) : -EBADF;
575 CERROR("lustre_lite: bad iget4 for root\n");
579 checksum = sbi->ll_flags & LL_SBI_CHECKSUM;
580 if (sbi->ll_checksum_set) {
581 err = obd_set_info_async(NULL, sbi->ll_dt_exp,
582 sizeof(KEY_CHECKSUM), KEY_CHECKSUM,
583 sizeof(checksum), &checksum, NULL);
585 CERROR("%s: Set checksum failed: rc = %d\n",
586 sbi->ll_dt_exp->exp_obd->obd_name, err);
592 sb->s_root = d_make_root(root);
593 if (sb->s_root == NULL) {
595 CERROR("%s: can't make root dentry: rc = %d\n",
596 sbi->ll_fsname, err);
599 #ifdef HAVE_DCACHE_LOCK
600 sb->s_root->d_op = &ll_d_ops;
603 sbi->ll_sdev_orig = sb->s_dev;
605 /* We set sb->s_dev equal on all lustre clients in order to support
606 * NFS export clustering. NFSD requires that the FSID be the same
608 /* s_dev is also used in lt_compare() to compare two fs, but that is
609 * only a node-local comparison. */
610 uuid = obd_get_uuid(sbi->ll_md_exp);
612 sb->s_dev = get_uuid2int(uuid->uuid, strlen(uuid->uuid));
619 if (sbi->ll_dt_obd) {
620 err = sysfs_create_link(&sbi->ll_kset.kobj,
621 &sbi->ll_dt_obd->obd_kset.kobj,
622 sbi->ll_dt_obd->obd_type->typ_name);
624 CERROR("%s: could not register %s in llite: rc = %d\n",
625 dt, sbi->ll_fsname, err);
630 if (sbi->ll_md_obd) {
631 err = sysfs_create_link(&sbi->ll_kset.kobj,
632 &sbi->ll_md_obd->obd_kset.kobj,
633 sbi->ll_md_obd->obd_type->typ_name);
635 CERROR("%s: could not register %s in llite: rc = %d\n",
636 md, sbi->ll_fsname, err);
646 obd_fid_fini(sbi->ll_dt_exp->exp_obd);
648 obd_disconnect(sbi->ll_dt_exp);
649 sbi->ll_dt_exp = NULL;
650 sbi->ll_dt_obd = NULL;
652 obd_fid_fini(sbi->ll_md_exp->exp_obd);
654 obd_disconnect(sbi->ll_md_exp);
655 sbi->ll_md_exp = NULL;
656 sbi->ll_md_obd = NULL;
665 int ll_get_max_mdsize(struct ll_sb_info *sbi, int *lmmsize)
669 size = sizeof(*lmmsize);
670 rc = obd_get_info(NULL, sbi->ll_dt_exp, sizeof(KEY_MAX_EASIZE),
671 KEY_MAX_EASIZE, &size, lmmsize);
673 CERROR("%s: cannot get max LOV EA size: rc = %d\n",
674 sbi->ll_dt_exp->exp_obd->obd_name, rc);
678 CDEBUG(D_INFO, "max LOV ea size: %d\n", *lmmsize);
681 rc = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_MAX_EASIZE),
682 KEY_MAX_EASIZE, &size, lmmsize);
684 CERROR("Get max mdsize error rc %d\n", rc);
686 CDEBUG(D_INFO, "max LMV ea size: %d\n", *lmmsize);
692 * Get the value of the default_easize parameter.
694 * \see client_obd::cl_default_mds_easize
696 * \param[in] sbi superblock info for this filesystem
697 * \param[out] lmmsize pointer to storage location for value
699 * \retval 0 on success
700 * \retval negative negated errno on failure
702 int ll_get_default_mdsize(struct ll_sb_info *sbi, int *lmmsize)
707 rc = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_DEFAULT_EASIZE),
708 KEY_DEFAULT_EASIZE, &size, lmmsize);
710 CERROR("Get default mdsize error rc %d\n", rc);
716 * Set the default_easize parameter to the given value.
718 * \see client_obd::cl_default_mds_easize
720 * \param[in] sbi superblock info for this filesystem
721 * \param[in] lmmsize the size to set
723 * \retval 0 on success
724 * \retval negative negated errno on failure
726 int ll_set_default_mdsize(struct ll_sb_info *sbi, int lmmsize)
730 if (lmmsize < sizeof(struct lov_mds_md) ||
731 lmmsize > OBD_MAX_DEFAULT_EA_SIZE)
734 rc = obd_set_info_async(NULL, sbi->ll_md_exp,
735 sizeof(KEY_DEFAULT_EASIZE), KEY_DEFAULT_EASIZE,
736 sizeof(int), &lmmsize, NULL);
741 static void client_common_put_super(struct super_block *sb)
743 struct ll_sb_info *sbi = ll_s2sbi(sb);
748 obd_fid_fini(sbi->ll_dt_exp->exp_obd);
749 obd_disconnect(sbi->ll_dt_exp);
750 sbi->ll_dt_exp = NULL;
752 ll_debugfs_unregister_super(sb);
754 obd_fid_fini(sbi->ll_md_exp->exp_obd);
755 obd_disconnect(sbi->ll_md_exp);
756 sbi->ll_md_exp = NULL;
761 void ll_kill_super(struct super_block *sb)
763 struct ll_sb_info *sbi;
767 if (!(sb->s_flags & MS_ACTIVE))
771 /* we need restore s_dev from changed for clustred NFS before put_super
772 * because new kernels have cached s_dev and change sb->s_dev in
773 * put_super not affected real removing devices */
775 sb->s_dev = sbi->ll_sdev_orig;
776 sbi->ll_umounting = 1;
778 /* wait running statahead threads to quit */
779 while (atomic_read(&sbi->ll_sa_running) > 0) {
780 set_current_state(TASK_UNINTERRUPTIBLE);
781 schedule_timeout(msecs_to_jiffies(MSEC_PER_SEC >> 3));
788 static inline int ll_set_opt(const char *opt, char *data, int fl)
790 if (strncmp(opt, data, strlen(opt)) != 0)
796 /* non-client-specific mount options are parsed in lmd_parse */
797 static int ll_options(char *options, struct ll_sb_info *sbi)
800 char *s1 = options, *s2;
801 int *flags = &sbi->ll_flags;
807 CDEBUG(D_CONFIG, "Parsing opts %s\n", options);
810 CDEBUG(D_SUPER, "next opt=%s\n", s1);
811 tmp = ll_set_opt("nolock", s1, LL_SBI_NOLCK);
816 tmp = ll_set_opt("flock", s1, LL_SBI_FLOCK);
821 tmp = ll_set_opt("localflock", s1, LL_SBI_LOCALFLOCK);
826 tmp = ll_set_opt("noflock", s1, LL_SBI_FLOCK|LL_SBI_LOCALFLOCK);
831 tmp = ll_set_opt("user_xattr", s1, LL_SBI_USER_XATTR);
836 tmp = ll_set_opt("nouser_xattr", s1, LL_SBI_USER_XATTR);
841 tmp = ll_set_opt("context", s1, 1);
844 tmp = ll_set_opt("fscontext", s1, 1);
847 tmp = ll_set_opt("defcontext", s1, 1);
850 tmp = ll_set_opt("rootcontext", s1, 1);
853 tmp = ll_set_opt("user_fid2path", s1, LL_SBI_USER_FID2PATH);
858 tmp = ll_set_opt("nouser_fid2path", s1, LL_SBI_USER_FID2PATH);
864 tmp = ll_set_opt("checksum", s1, LL_SBI_CHECKSUM);
867 sbi->ll_checksum_set = 1;
870 tmp = ll_set_opt("nochecksum", s1, LL_SBI_CHECKSUM);
873 sbi->ll_checksum_set = 1;
876 tmp = ll_set_opt("lruresize", s1, LL_SBI_LRU_RESIZE);
881 tmp = ll_set_opt("nolruresize", s1, LL_SBI_LRU_RESIZE);
886 tmp = ll_set_opt("lazystatfs", s1, LL_SBI_LAZYSTATFS);
891 tmp = ll_set_opt("nolazystatfs", s1, LL_SBI_LAZYSTATFS);
896 tmp = ll_set_opt("32bitapi", s1, LL_SBI_32BIT_API);
901 tmp = ll_set_opt("verbose", s1, LL_SBI_VERBOSE);
906 tmp = ll_set_opt("noverbose", s1, LL_SBI_VERBOSE);
911 tmp = ll_set_opt("always_ping", s1, LL_SBI_ALWAYS_PING);
916 LCONSOLE_ERROR_MSG(0x152, "Unknown option '%s', won't mount.\n",
922 s2 = strchr(s1, ',');
930 void ll_lli_init(struct ll_inode_info *lli)
932 lli->lli_inode_magic = LLI_INODE_MAGIC;
934 spin_lock_init(&lli->lli_lock);
935 lli->lli_posix_acl = NULL;
936 /* Do not set lli_fid, it has been initialized already. */
937 fid_zero(&lli->lli_pfid);
938 lli->lli_mds_read_och = NULL;
939 lli->lli_mds_write_och = NULL;
940 lli->lli_mds_exec_och = NULL;
941 lli->lli_open_fd_read_count = 0;
942 lli->lli_open_fd_write_count = 0;
943 lli->lli_open_fd_exec_count = 0;
944 mutex_init(&lli->lli_och_mutex);
945 spin_lock_init(&lli->lli_agl_lock);
946 spin_lock_init(&lli->lli_layout_lock);
947 ll_layout_version_set(lli, CL_LAYOUT_GEN_NONE);
948 lli->lli_clob = NULL;
950 init_rwsem(&lli->lli_xattrs_list_rwsem);
951 mutex_init(&lli->lli_xattrs_enq_lock);
953 LASSERT(lli->lli_vfs_inode.i_mode != 0);
954 if (S_ISDIR(lli->lli_vfs_inode.i_mode)) {
955 mutex_init(&lli->lli_readdir_mutex);
956 lli->lli_opendir_key = NULL;
958 spin_lock_init(&lli->lli_sa_lock);
959 lli->lli_opendir_pid = 0;
960 lli->lli_sa_enabled = 0;
961 lli->lli_def_stripe_offset = -1;
962 init_rwsem(&lli->lli_lsm_sem);
964 mutex_init(&lli->lli_size_mutex);
965 lli->lli_symlink_name = NULL;
966 init_rwsem(&lli->lli_trunc_sem);
967 range_lock_tree_init(&lli->lli_write_tree);
968 init_rwsem(&lli->lli_glimpse_sem);
969 lli->lli_glimpse_time = ktime_set(0, 0);
970 INIT_LIST_HEAD(&lli->lli_agl_list);
971 lli->lli_agl_index = 0;
972 lli->lli_async_rc = 0;
973 spin_lock_init(&lli->lli_heat_lock);
974 obd_heat_clear(lli->lli_heat_instances, OBD_HEAT_COUNT);
975 lli->lli_heat_flags = 0;
977 mutex_init(&lli->lli_layout_mutex);
978 memset(lli->lli_jobid, 0, sizeof(lli->lli_jobid));
981 #define MAX_STRING_SIZE 128
983 #ifndef HAVE_SUPER_SETUP_BDI_NAME
985 #define LSI_BDI_INITIALIZED 0x00400000
987 #ifndef HAVE_BDI_CAP_MAP_COPY
988 # define BDI_CAP_MAP_COPY 0
991 static int super_setup_bdi_name(struct super_block *sb, char *fmt, ...)
993 struct lustre_sb_info *lsi = s2lsi(sb);
994 char buf[MAX_STRING_SIZE];
998 err = bdi_init(&lsi->lsi_bdi);
1002 lsi->lsi_flags |= LSI_BDI_INITIALIZED;
1003 lsi->lsi_bdi.capabilities = BDI_CAP_MAP_COPY;
1004 lsi->lsi_bdi.name = "lustre";
1005 va_start(args, fmt);
1006 vsnprintf(buf, MAX_STRING_SIZE, fmt, args);
1008 err = bdi_register(&lsi->lsi_bdi, NULL, "%s", buf);
1011 sb->s_bdi = &lsi->lsi_bdi;
1015 #endif /* !HAVE_SUPER_SETUP_BDI_NAME */
1017 int ll_fill_super(struct super_block *sb, struct vfsmount *mnt)
1019 struct lustre_profile *lprof = NULL;
1020 struct lustre_sb_info *lsi = s2lsi(sb);
1021 struct ll_sb_info *sbi = NULL;
1022 char *dt = NULL, *md = NULL;
1023 char *profilenm = get_profile_name(sb);
1024 struct config_llog_instance *cfg;
1025 /* %p for void* in printf needs 16+2 characters: 0xffffffffffffffff */
1026 const int instlen = 16 + 2;
1027 unsigned long cfg_instance = ll_get_cfg_instance(sb);
1028 char name[MAX_STRING_SIZE];
1037 /* for ASLR, to map between cfg_instance and hashed ptr */
1038 CDEBUG(D_VFSTRACE, "VFS Op: cfg_instance %s-%016lx (sb %p)\n",
1039 profilenm, cfg_instance, sb);
1041 try_module_get(THIS_MODULE);
1045 GOTO(out_free_cfg, err = -ENOMEM);
1047 /* client additional sb info */
1048 lsi->lsi_llsbi = sbi = ll_init_sbi();
1050 GOTO(out_free_cfg, err = -ENOMEM);
1052 err = ll_options(lsi->lsi_lmd->lmd_opts, sbi);
1054 GOTO(out_free_cfg, err);
1056 #ifndef HAVE_DCACHE_LOCK
1057 /* kernel >= 2.6.38 store dentry operations in sb->s_d_op. */
1058 sb->s_d_op = &ll_d_ops;
1061 generate_random_uuid(uuid.b);
1062 snprintf(sbi->ll_sb_uuid.uuid, UUID_SIZE, "%pU", uuid.b);
1064 CDEBUG(D_CONFIG, "llite sb uuid: %s\n", sbi->ll_sb_uuid.uuid);
1067 len = strlen(profilenm);
1068 ptr = strrchr(profilenm, '-');
1069 if (ptr && (strcmp(ptr, "-client") == 0))
1072 if (len > LUSTRE_MAXFSNAME) {
1073 if (unlikely(len >= MAX_STRING_SIZE))
1074 len = MAX_STRING_SIZE - 1;
1075 strncpy(name, profilenm, len);
1077 err = -ENAMETOOLONG;
1078 CERROR("%s: fsname longer than %u characters: rc = %d\n",
1079 name, LUSTRE_MAXFSNAME, err);
1080 GOTO(out_free_cfg, err);
1082 strncpy(sbi->ll_fsname, profilenm, len);
1083 sbi->ll_fsname[len] = '\0';
1086 snprintf(name, sizeof(name), "%.*s-%016lx", len,
1087 profilenm, cfg_instance);
1089 err = super_setup_bdi_name(sb, "%s", name);
1091 GOTO(out_free_cfg, err);
1093 /* Call ll_debugfs_register_super() before lustre_process_log()
1094 * so that "llite.*.*" params can be processed correctly.
1096 err = ll_debugfs_register_super(sb, name);
1098 CERROR("%s: could not register mountpoint in llite: rc = %d\n",
1099 sbi->ll_fsname, err);
1103 /* The cfg_instance is a value unique to this super, in case some
1104 * joker tries to mount the same fs at two mount points.
1106 cfg->cfg_instance = cfg_instance;
1107 cfg->cfg_uuid = lsi->lsi_llsbi->ll_sb_uuid;
1108 cfg->cfg_callback = class_config_llog_handler;
1109 cfg->cfg_sub_clds = CONFIG_SUB_CLIENT;
1110 /* set up client obds */
1111 err = lustre_process_log(sb, profilenm, cfg);
1113 GOTO(out_debugfs, err);
1115 /* Profile set with LCFG_MOUNTOPT so we can find our mdc and osc obds */
1116 lprof = class_get_profile(profilenm);
1117 if (lprof == NULL) {
1118 LCONSOLE_ERROR_MSG(0x156, "The client profile '%s' could not be"
1119 " read from the MGS. Does that filesystem "
1120 "exist?\n", profilenm);
1121 GOTO(out_debugfs, err = -EINVAL);
1123 CDEBUG(D_CONFIG, "Found profile %s: mdc=%s osc=%s\n", profilenm,
1124 lprof->lp_md, lprof->lp_dt);
1126 dt_len = strlen(lprof->lp_dt) + instlen + 2;
1127 OBD_ALLOC(dt, dt_len);
1129 GOTO(out_profile, err = -ENOMEM);
1130 snprintf(dt, dt_len - 1, "%s-%016lx", lprof->lp_dt, cfg_instance);
1132 md_len = strlen(lprof->lp_md) + instlen + 2;
1133 OBD_ALLOC(md, md_len);
1135 GOTO(out_free_dt, err = -ENOMEM);
1136 snprintf(md, md_len - 1, "%s-%016lx", lprof->lp_md, cfg_instance);
1138 /* connections, registrations, sb setup */
1139 err = client_common_fill_super(sb, md, dt, mnt);
1141 GOTO(out_free_md, err);
1143 sbi->ll_client_common_fill_super_succeeded = 1;
1147 OBD_FREE(md, md_len);
1150 OBD_FREE(dt, dt_len);
1153 class_put_profile(lprof);
1156 ll_debugfs_unregister_super(sb);
1163 else if (sbi->ll_flags & LL_SBI_VERBOSE)
1164 LCONSOLE_WARN("Mounted %s\n", profilenm);
1166 } /* ll_fill_super */
1168 void ll_put_super(struct super_block *sb)
1170 struct config_llog_instance cfg, params_cfg;
1171 struct obd_device *obd;
1172 struct lustre_sb_info *lsi = s2lsi(sb);
1173 struct ll_sb_info *sbi = ll_s2sbi(sb);
1174 char *profilenm = get_profile_name(sb);
1175 unsigned long cfg_instance = ll_get_cfg_instance(sb);
1177 int next, force = 1, rc = 0;
1181 GOTO(out_no_sbi, 0);
1183 /* Should replace instance_id with something better for ASLR */
1184 CDEBUG(D_VFSTRACE, "VFS Op: cfg_instance %s-%016lx (sb %p)\n",
1185 profilenm, cfg_instance, sb);
1187 cfg.cfg_instance = cfg_instance;
1188 lustre_end_log(sb, profilenm, &cfg);
1190 params_cfg.cfg_instance = cfg_instance;
1191 lustre_end_log(sb, PARAMS_FILENAME, ¶ms_cfg);
1193 if (sbi->ll_md_exp) {
1194 obd = class_exp2obd(sbi->ll_md_exp);
1196 force = obd->obd_force;
1199 /* Wait for unstable pages to be committed to stable storage */
1201 struct l_wait_info lwi = LWI_INTR(LWI_ON_SIGNAL_NOOP, NULL);
1202 rc = l_wait_event(sbi->ll_cache->ccc_unstable_waitq,
1203 atomic_long_read(&sbi->ll_cache->ccc_unstable_nr) == 0,
1207 ccc_count = atomic_long_read(&sbi->ll_cache->ccc_unstable_nr);
1208 if (force == 0 && rc != -EINTR)
1209 LASSERTF(ccc_count == 0, "count: %li\n", ccc_count);
1211 /* We need to set force before the lov_disconnect in
1212 lustre_common_put_super, since l_d cleans up osc's as well. */
1215 while ((obd = class_devices_in_group(&sbi->ll_sb_uuid,
1217 obd->obd_force = force;
1221 if (sbi->ll_client_common_fill_super_succeeded) {
1222 /* Only if client_common_fill_super succeeded */
1223 client_common_put_super(sb);
1227 while ((obd = class_devices_in_group(&sbi->ll_sb_uuid, &next)) !=NULL) {
1228 class_manual_cleanup(obd);
1231 if (sbi->ll_flags & LL_SBI_VERBOSE)
1232 LCONSOLE_WARN("Unmounted %s\n", profilenm ? profilenm : "");
1235 class_del_profile(profilenm);
1237 #ifndef HAVE_SUPER_SETUP_BDI_NAME
1238 if (lsi->lsi_flags & LSI_BDI_INITIALIZED) {
1239 bdi_destroy(&lsi->lsi_bdi);
1240 lsi->lsi_flags &= ~LSI_BDI_INITIALIZED;
1245 lsi->lsi_llsbi = NULL;
1247 lustre_common_put_super(sb);
1249 cl_env_cache_purge(~0);
1251 module_put(THIS_MODULE);
1254 } /* client_put_super */
1256 struct inode *ll_inode_from_resource_lock(struct ldlm_lock *lock)
1258 struct inode *inode = NULL;
1260 /* NOTE: we depend on atomic igrab() -bzzz */
1261 lock_res_and_lock(lock);
1262 if (lock->l_resource->lr_lvb_inode) {
1263 struct ll_inode_info * lli;
1264 lli = ll_i2info(lock->l_resource->lr_lvb_inode);
1265 if (lli->lli_inode_magic == LLI_INODE_MAGIC) {
1266 inode = igrab(lock->l_resource->lr_lvb_inode);
1268 inode = lock->l_resource->lr_lvb_inode;
1269 LDLM_DEBUG_LIMIT(inode->i_state & I_FREEING ? D_INFO :
1270 D_WARNING, lock, "lr_lvb_inode %p is "
1271 "bogus: magic %08x",
1272 lock->l_resource->lr_lvb_inode,
1273 lli->lli_inode_magic);
1277 unlock_res_and_lock(lock);
1281 void ll_dir_clear_lsm_md(struct inode *inode)
1283 struct ll_inode_info *lli = ll_i2info(inode);
1285 LASSERT(S_ISDIR(inode->i_mode));
1287 if (lli->lli_lsm_md != NULL) {
1288 lmv_free_memmd(lli->lli_lsm_md);
1289 lli->lli_lsm_md = NULL;
1293 static struct inode *ll_iget_anon_dir(struct super_block *sb,
1294 const struct lu_fid *fid,
1295 struct lustre_md *md)
1297 struct ll_sb_info *sbi = ll_s2sbi(sb);
1298 struct mdt_body *body = md->body;
1299 struct inode *inode;
1303 ino = cl_fid_build_ino(fid, sbi->ll_flags & LL_SBI_32BIT_API);
1304 inode = iget_locked(sb, ino);
1305 if (inode == NULL) {
1306 CERROR("%s: failed get simple inode "DFID": rc = -ENOENT\n",
1307 sbi->ll_fsname, PFID(fid));
1308 RETURN(ERR_PTR(-ENOENT));
1311 if (inode->i_state & I_NEW) {
1312 struct ll_inode_info *lli = ll_i2info(inode);
1313 struct lmv_stripe_md *lsm = md->lmv;
1315 inode->i_mode = (inode->i_mode & ~S_IFMT) |
1316 (body->mbo_mode & S_IFMT);
1317 LASSERTF(S_ISDIR(inode->i_mode), "Not slave inode "DFID"\n",
1320 inode->i_mtime.tv_sec = 0;
1321 inode->i_atime.tv_sec = 0;
1322 inode->i_ctime.tv_sec = 0;
1325 #ifdef HAVE_BACKING_DEV_INFO
1326 /* initializing backing dev info. */
1327 inode->i_mapping->backing_dev_info =
1328 &s2lsi(inode->i_sb)->lsi_bdi;
1330 inode->i_op = &ll_dir_inode_operations;
1331 inode->i_fop = &ll_dir_operations;
1332 lli->lli_fid = *fid;
1335 LASSERT(lsm != NULL);
1336 /* master object FID */
1337 lli->lli_pfid = body->mbo_fid1;
1338 CDEBUG(D_INODE, "lli %p slave "DFID" master "DFID"\n",
1339 lli, PFID(fid), PFID(&lli->lli_pfid));
1340 unlock_new_inode(inode);
1346 static int ll_init_lsm_md(struct inode *inode, struct lustre_md *md)
1349 struct lmv_stripe_md *lsm = md->lmv;
1350 struct ll_inode_info *lli = ll_i2info(inode);
1353 LASSERT(lsm != NULL);
1355 CDEBUG(D_INODE, "%s: "DFID" set dir layout:\n",
1356 ll_i2sbi(inode)->ll_fsname, PFID(&lli->lli_fid));
1357 lsm_md_dump(D_INODE, lsm);
1359 /* XXX sigh, this lsm_root initialization should be in
1360 * LMV layer, but it needs ll_iget right now, so we
1361 * put this here right now. */
1362 for (i = 0; i < lsm->lsm_md_stripe_count; i++) {
1363 fid = &lsm->lsm_md_oinfo[i].lmo_fid;
1364 LASSERT(lsm->lsm_md_oinfo[i].lmo_root == NULL);
1365 /* Unfortunately ll_iget will call ll_update_inode,
1366 * where the initialization of slave inode is slightly
1367 * different, so it reset lsm_md to NULL to avoid
1368 * initializing lsm for slave inode. */
1369 lsm->lsm_md_oinfo[i].lmo_root =
1370 ll_iget_anon_dir(inode->i_sb, fid, md);
1371 if (IS_ERR(lsm->lsm_md_oinfo[i].lmo_root)) {
1372 int rc = PTR_ERR(lsm->lsm_md_oinfo[i].lmo_root);
1374 lsm->lsm_md_oinfo[i].lmo_root = NULL;
1376 iput(lsm->lsm_md_oinfo[i].lmo_root);
1377 lsm->lsm_md_oinfo[i].lmo_root = NULL;
1383 lli->lli_lsm_md = lsm;
1388 static int ll_update_lsm_md(struct inode *inode, struct lustre_md *md)
1390 struct ll_inode_info *lli = ll_i2info(inode);
1391 struct lmv_stripe_md *lsm = md->lmv;
1396 LASSERT(S_ISDIR(inode->i_mode));
1397 CDEBUG(D_INODE, "update lsm %p of "DFID"\n", lli->lli_lsm_md,
1398 PFID(ll_inode2fid(inode)));
1401 * no striped information from request, lustre_md from req does not
1402 * include stripeEA, see ll_md_setattr()
1408 * normally dir layout doesn't change, only take read lock to check
1409 * that to avoid blocking other MD operations.
1411 if (lli->lli_lsm_md)
1412 down_read(&lli->lli_lsm_sem);
1414 down_write(&lli->lli_lsm_sem);
1417 * if dir layout mismatch, check whether version is increased, which
1418 * means layout is changed, this happens in dir migration and lfsck.
1420 * foreign LMV should not change.
1422 if (lli->lli_lsm_md &&
1423 lli->lli_lsm_md->lsm_md_magic != LMV_MAGIC_FOREIGN &&
1424 !lsm_md_eq(lli->lli_lsm_md, lsm)) {
1425 if (lsm->lsm_md_layout_version <=
1426 lli->lli_lsm_md->lsm_md_layout_version) {
1427 CERROR("%s: "DFID" dir layout mismatch:\n",
1428 ll_i2sbi(inode)->ll_fsname,
1429 PFID(&lli->lli_fid));
1430 lsm_md_dump(D_ERROR, lli->lli_lsm_md);
1431 lsm_md_dump(D_ERROR, lsm);
1432 GOTO(unlock, rc = -EINVAL);
1435 /* layout changed, switch to write lock */
1436 up_read(&lli->lli_lsm_sem);
1437 down_write(&lli->lli_lsm_sem);
1438 ll_dir_clear_lsm_md(inode);
1441 /* set directory layout */
1442 if (!lli->lli_lsm_md) {
1443 struct cl_attr *attr;
1445 if (lsm->lsm_md_magic == LMV_MAGIC_FOREIGN) {
1446 /* set md->lmv to NULL, so the following free lustre_md
1447 * will not free this lsm */
1449 lli->lli_lsm_md = lsm;
1450 up_write(&lli->lli_lsm_sem);
1454 rc = ll_init_lsm_md(inode, md);
1455 up_write(&lli->lli_lsm_sem);
1459 /* set md->lmv to NULL, so the following free lustre_md
1460 * will not free this lsm */
1464 * md_merge_attr() may take long, since lsm is already set,
1465 * switch to read lock.
1467 down_read(&lli->lli_lsm_sem);
1469 OBD_ALLOC_PTR(attr);
1471 GOTO(unlock, rc = -ENOMEM);
1473 /* validate the lsm */
1474 rc = md_merge_attr(ll_i2mdexp(inode), lsm, attr,
1475 ll_md_blocking_ast);
1481 if (md->body->mbo_valid & OBD_MD_FLNLINK)
1482 md->body->mbo_nlink = attr->cat_nlink;
1483 if (md->body->mbo_valid & OBD_MD_FLSIZE)
1484 md->body->mbo_size = attr->cat_size;
1485 if (md->body->mbo_valid & OBD_MD_FLATIME)
1486 md->body->mbo_atime = attr->cat_atime;
1487 if (md->body->mbo_valid & OBD_MD_FLCTIME)
1488 md->body->mbo_ctime = attr->cat_ctime;
1489 if (md->body->mbo_valid & OBD_MD_FLMTIME)
1490 md->body->mbo_mtime = attr->cat_mtime;
1495 up_read(&lli->lli_lsm_sem);
1500 void ll_clear_inode(struct inode *inode)
1502 struct ll_inode_info *lli = ll_i2info(inode);
1503 struct ll_sb_info *sbi = ll_i2sbi(inode);
1506 CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p)\n",
1507 PFID(ll_inode2fid(inode)), inode);
1509 if (S_ISDIR(inode->i_mode)) {
1510 /* these should have been cleared in ll_file_release */
1511 LASSERT(lli->lli_opendir_key == NULL);
1512 LASSERT(lli->lli_sai == NULL);
1513 LASSERT(lli->lli_opendir_pid == 0);
1516 md_null_inode(sbi->ll_md_exp, ll_inode2fid(inode));
1518 LASSERT(!lli->lli_open_fd_write_count);
1519 LASSERT(!lli->lli_open_fd_read_count);
1520 LASSERT(!lli->lli_open_fd_exec_count);
1522 if (lli->lli_mds_write_och)
1523 ll_md_real_close(inode, FMODE_WRITE);
1524 if (lli->lli_mds_exec_och)
1525 ll_md_real_close(inode, FMODE_EXEC);
1526 if (lli->lli_mds_read_och)
1527 ll_md_real_close(inode, FMODE_READ);
1529 if (S_ISLNK(inode->i_mode) && lli->lli_symlink_name) {
1530 OBD_FREE(lli->lli_symlink_name,
1531 strlen(lli->lli_symlink_name) + 1);
1532 lli->lli_symlink_name = NULL;
1535 ll_xattr_cache_destroy(inode);
1537 #ifdef CONFIG_FS_POSIX_ACL
1538 forget_all_cached_acls(inode);
1539 if (lli->lli_posix_acl) {
1540 posix_acl_release(lli->lli_posix_acl);
1541 lli->lli_posix_acl = NULL;
1544 lli->lli_inode_magic = LLI_INODE_DEAD;
1546 if (S_ISDIR(inode->i_mode))
1547 ll_dir_clear_lsm_md(inode);
1548 else if (S_ISREG(inode->i_mode) && !is_bad_inode(inode))
1549 LASSERT(list_empty(&lli->lli_agl_list));
1552 * XXX This has to be done before lsm is freed below, because
1553 * cl_object still uses inode lsm.
1555 cl_inode_fini(inode);
1560 static int ll_md_setattr(struct dentry *dentry, struct md_op_data *op_data)
1562 struct lustre_md md;
1563 struct inode *inode = dentry->d_inode;
1564 struct ll_sb_info *sbi = ll_i2sbi(inode);
1565 struct ptlrpc_request *request = NULL;
1569 op_data = ll_prep_md_op_data(op_data, inode, NULL, NULL, 0, 0,
1570 LUSTRE_OPC_ANY, NULL);
1571 if (IS_ERR(op_data))
1572 RETURN(PTR_ERR(op_data));
1574 rc = md_setattr(sbi->ll_md_exp, op_data, NULL, 0, &request);
1576 ptlrpc_req_finished(request);
1577 if (rc == -ENOENT) {
1579 /* Unlinked special device node? Or just a race?
1580 * Pretend we done everything. */
1581 if (!S_ISREG(inode->i_mode) &&
1582 !S_ISDIR(inode->i_mode)) {
1583 ia_valid = op_data->op_attr.ia_valid;
1584 op_data->op_attr.ia_valid &= ~TIMES_SET_FLAGS;
1585 rc = simple_setattr(dentry, &op_data->op_attr);
1586 op_data->op_attr.ia_valid = ia_valid;
1588 } else if (rc != -EPERM && rc != -EACCES && rc != -ETXTBSY) {
1589 CERROR("md_setattr fails: rc = %d\n", rc);
1594 rc = md_get_lustre_md(sbi->ll_md_exp, request, sbi->ll_dt_exp,
1595 sbi->ll_md_exp, &md);
1597 ptlrpc_req_finished(request);
1601 ia_valid = op_data->op_attr.ia_valid;
1602 /* inode size will be in ll_setattr_ost, can't do it now since dirty
1603 * cache is not cleared yet. */
1604 op_data->op_attr.ia_valid &= ~(TIMES_SET_FLAGS | ATTR_SIZE);
1605 if (S_ISREG(inode->i_mode))
1607 rc = simple_setattr(dentry, &op_data->op_attr);
1608 if (S_ISREG(inode->i_mode))
1609 inode_unlock(inode);
1610 op_data->op_attr.ia_valid = ia_valid;
1612 rc = ll_update_inode(inode, &md);
1613 ptlrpc_req_finished(request);
1618 /* If this inode has objects allocated to it (lsm != NULL), then the OST
1619 * object(s) determine the file size and mtime. Otherwise, the MDS will
1620 * keep these values until such a time that objects are allocated for it.
1621 * We do the MDS operations first, as it is checking permissions for us.
1622 * We don't to the MDS RPC if there is nothing that we want to store there,
1623 * otherwise there is no harm in updating mtime/atime on the MDS if we are
1624 * going to do an RPC anyways.
1626 * If we are doing a truncate, we will send the mtime and ctime updates
1627 * to the OST with the punch RPC, otherwise we do an explicit setattr RPC.
1628 * I don't believe it is possible to get e.g. ATTR_MTIME_SET and ATTR_SIZE
1631 * In case of HSMimport, we only set attr on MDS.
1633 int ll_setattr_raw(struct dentry *dentry, struct iattr *attr,
1634 enum op_xvalid xvalid, bool hsm_import)
1636 struct inode *inode = dentry->d_inode;
1637 struct ll_inode_info *lli = ll_i2info(inode);
1638 struct md_op_data *op_data = NULL;
1642 CDEBUG(D_VFSTRACE, "%s: setattr inode "DFID"(%p) from %llu to %llu, "
1643 "valid %x, hsm_import %d\n",
1644 ll_i2sbi(inode)->ll_fsname, PFID(&lli->lli_fid),
1645 inode, i_size_read(inode), attr->ia_size, attr->ia_valid,
1648 if (attr->ia_valid & ATTR_SIZE) {
1649 /* Check new size against VFS/VM file size limit and rlimit */
1650 rc = inode_newsize_ok(inode, attr->ia_size);
1654 /* The maximum Lustre file size is variable, based on the
1655 * OST maximum object size and number of stripes. This
1656 * needs another check in addition to the VFS check above. */
1657 if (attr->ia_size > ll_file_maxbytes(inode)) {
1658 CDEBUG(D_INODE,"file "DFID" too large %llu > %llu\n",
1659 PFID(&lli->lli_fid), attr->ia_size,
1660 ll_file_maxbytes(inode));
1664 attr->ia_valid |= ATTR_MTIME | ATTR_CTIME;
1667 /* POSIX: check before ATTR_*TIME_SET set (from inode_change_ok) */
1668 if (attr->ia_valid & TIMES_SET_FLAGS) {
1669 if ((!uid_eq(current_fsuid(), inode->i_uid)) &&
1670 !cfs_capable(CFS_CAP_FOWNER))
1674 /* We mark all of the fields "set" so MDS/OST does not re-set them */
1675 if (!(xvalid & OP_XVALID_CTIME_SET) &&
1676 (attr->ia_valid & ATTR_CTIME)) {
1677 attr->ia_ctime = current_time(inode);
1678 xvalid |= OP_XVALID_CTIME_SET;
1680 if (!(attr->ia_valid & ATTR_ATIME_SET) &&
1681 (attr->ia_valid & ATTR_ATIME)) {
1682 attr->ia_atime = current_time(inode);
1683 attr->ia_valid |= ATTR_ATIME_SET;
1685 if (!(attr->ia_valid & ATTR_MTIME_SET) &&
1686 (attr->ia_valid & ATTR_MTIME)) {
1687 attr->ia_mtime = current_time(inode);
1688 attr->ia_valid |= ATTR_MTIME_SET;
1691 if (attr->ia_valid & (ATTR_MTIME | ATTR_CTIME))
1692 CDEBUG(D_INODE, "setting mtime %lld, ctime %lld, now = %lld\n",
1693 (s64)attr->ia_mtime.tv_sec, (s64)attr->ia_ctime.tv_sec,
1694 ktime_get_real_seconds());
1696 if (S_ISREG(inode->i_mode)) {
1697 if (attr->ia_valid & ATTR_SIZE)
1698 inode_dio_write_done(inode);
1699 inode_unlock(inode);
1702 /* We always do an MDS RPC, even if we're only changing the size;
1703 * only the MDS knows whether truncate() should fail with -ETXTBUSY */
1705 OBD_ALLOC_PTR(op_data);
1706 if (op_data == NULL)
1707 GOTO(out, rc = -ENOMEM);
1709 if (!hsm_import && attr->ia_valid & ATTR_SIZE) {
1710 /* If we are changing file size, file content is
1711 * modified, flag it.
1713 xvalid |= OP_XVALID_OWNEROVERRIDE;
1714 op_data->op_bias |= MDS_DATA_MODIFIED;
1715 ll_file_clear_flag(lli, LLIF_DATA_MODIFIED);
1718 if (attr->ia_valid & ATTR_FILE) {
1719 struct ll_file_data *fd = LUSTRE_FPRIVATE(attr->ia_file);
1721 if (fd->fd_lease_och)
1722 op_data->op_bias |= MDS_TRUNC_KEEP_LEASE;
1725 op_data->op_attr = *attr;
1726 op_data->op_xvalid = xvalid;
1728 rc = ll_md_setattr(dentry, op_data);
1732 if (!S_ISREG(inode->i_mode) || hsm_import)
1735 if (attr->ia_valid & (ATTR_SIZE | ATTR_ATIME | ATTR_ATIME_SET |
1736 ATTR_MTIME | ATTR_MTIME_SET | ATTR_CTIME) ||
1737 xvalid & OP_XVALID_CTIME_SET) {
1738 /* For truncate and utimes sending attributes to OSTs, setting
1739 * mtime/atime to the past will be performed under PW [0:EOF]
1740 * extent lock (new_size:EOF for truncate). It may seem
1741 * excessive to send mtime/atime updates to OSTs when not
1742 * setting times to past, but it is necessary due to possible
1743 * time de-synchronization between MDT inode and OST objects
1745 rc = cl_setattr_ost(lli->lli_clob, attr, xvalid, 0);
1748 /* If the file was restored, it needs to set dirty flag.
1750 * We've already sent MDS_DATA_MODIFIED flag in
1751 * ll_md_setattr() for truncate. However, the MDT refuses to
1752 * set the HS_DIRTY flag on released files, so we have to set
1753 * it again if the file has been restored. Please check how
1754 * LLIF_DATA_MODIFIED is set in vvp_io_setattr_fini().
1756 * Please notice that if the file is not released, the previous
1757 * MDS_DATA_MODIFIED has taken effect and usually
1758 * LLIF_DATA_MODIFIED is not set(see vvp_io_setattr_fini()).
1759 * This way we can save an RPC for common open + trunc
1761 if (ll_file_test_and_clear_flag(lli, LLIF_DATA_MODIFIED)) {
1762 struct hsm_state_set hss = {
1763 .hss_valid = HSS_SETMASK,
1764 .hss_setmask = HS_DIRTY,
1768 rc2 = ll_hsm_state_set(inode, &hss);
1769 /* truncate and write can happen at the same time, so that
1770 * the file can be set modified even though the file is not
1771 * restored from released state, and ll_hsm_state_set() is
1772 * not applicable for the file, and rc2 < 0 is normal in this
1775 CDEBUG(D_INFO, DFID "HSM set dirty failed: rc2 = %d\n",
1776 PFID(ll_inode2fid(inode)), rc2);
1781 if (op_data != NULL)
1782 ll_finish_md_op_data(op_data);
1784 if (S_ISREG(inode->i_mode)) {
1786 if ((attr->ia_valid & ATTR_SIZE) && !hsm_import)
1787 inode_dio_wait(inode);
1788 /* Once we've got the i_mutex, it's safe to set the S_NOSEC
1789 * flag. ll_update_inode (called from ll_md_setattr), clears
1790 * inode flags, so there is a gap where S_NOSEC is not set.
1791 * This can cause a writer to take the i_mutex unnecessarily,
1792 * but this is safe to do and should be rare. */
1793 inode_has_no_xattr(inode);
1796 ll_stats_ops_tally(ll_i2sbi(inode), (attr->ia_valid & ATTR_SIZE) ?
1797 LPROC_LL_TRUNC : LPROC_LL_SETATTR, 1);
1802 int ll_setattr(struct dentry *de, struct iattr *attr)
1804 int mode = de->d_inode->i_mode;
1805 enum op_xvalid xvalid = 0;
1807 if ((attr->ia_valid & (ATTR_CTIME|ATTR_SIZE|ATTR_MODE)) ==
1808 (ATTR_CTIME|ATTR_SIZE|ATTR_MODE))
1809 xvalid |= OP_XVALID_OWNEROVERRIDE;
1811 if (((attr->ia_valid & (ATTR_MODE|ATTR_FORCE|ATTR_SIZE)) ==
1812 (ATTR_SIZE|ATTR_MODE)) &&
1813 (((mode & S_ISUID) && !(attr->ia_mode & S_ISUID)) ||
1814 (((mode & (S_ISGID|S_IXGRP)) == (S_ISGID|S_IXGRP)) &&
1815 !(attr->ia_mode & S_ISGID))))
1816 attr->ia_valid |= ATTR_FORCE;
1818 if ((attr->ia_valid & ATTR_MODE) &&
1820 !(attr->ia_mode & S_ISUID) &&
1821 !(attr->ia_valid & ATTR_KILL_SUID))
1822 attr->ia_valid |= ATTR_KILL_SUID;
1824 if ((attr->ia_valid & ATTR_MODE) &&
1825 ((mode & (S_ISGID|S_IXGRP)) == (S_ISGID|S_IXGRP)) &&
1826 !(attr->ia_mode & S_ISGID) &&
1827 !(attr->ia_valid & ATTR_KILL_SGID))
1828 attr->ia_valid |= ATTR_KILL_SGID;
1830 return ll_setattr_raw(de, attr, xvalid, false);
1833 int ll_statfs_internal(struct ll_sb_info *sbi, struct obd_statfs *osfs,
1836 struct obd_statfs obd_osfs = { 0 };
1841 max_age = ktime_get_seconds() - OBD_STATFS_CACHE_SECONDS;
1843 rc = obd_statfs(NULL, sbi->ll_md_exp, osfs, max_age, flags);
1847 osfs->os_type = LL_SUPER_MAGIC;
1849 CDEBUG(D_SUPER, "MDC blocks %llu/%llu objects %llu/%llu\n",
1850 osfs->os_bavail, osfs->os_blocks, osfs->os_ffree, osfs->os_files);
1852 if (osfs->os_state & OS_STATE_SUM)
1855 if (sbi->ll_flags & LL_SBI_LAZYSTATFS)
1856 flags |= OBD_STATFS_NODELAY;
1858 rc = obd_statfs(NULL, sbi->ll_dt_exp, &obd_osfs, max_age, flags);
1859 if (rc) /* Possibly a filesystem with no OSTs. Report MDT totals. */
1862 CDEBUG(D_SUPER, "OSC blocks %llu/%llu objects %llu/%llu\n",
1863 obd_osfs.os_bavail, obd_osfs.os_blocks, obd_osfs.os_ffree,
1866 osfs->os_bsize = obd_osfs.os_bsize;
1867 osfs->os_blocks = obd_osfs.os_blocks;
1868 osfs->os_bfree = obd_osfs.os_bfree;
1869 osfs->os_bavail = obd_osfs.os_bavail;
1871 /* If we have _some_ OSTs, but don't have as many free objects on the
1872 * OSTs as inodes on the MDTs, reduce the reported number of inodes
1873 * to compensate, so that the "inodes in use" number is correct.
1874 * This should be kept in sync with lod_statfs() behaviour.
1876 if (obd_osfs.os_files && obd_osfs.os_ffree < osfs->os_ffree) {
1877 osfs->os_files = (osfs->os_files - osfs->os_ffree) +
1879 osfs->os_ffree = obd_osfs.os_ffree;
1885 int ll_statfs(struct dentry *de, struct kstatfs *sfs)
1887 struct super_block *sb = de->d_sb;
1888 struct obd_statfs osfs;
1889 __u64 fsid = huge_encode_dev(sb->s_dev);
1892 CDEBUG(D_VFSTRACE, "VFS Op: at %llu jiffies\n", get_jiffies_64());
1893 ll_stats_ops_tally(ll_s2sbi(sb), LPROC_LL_STAFS, 1);
1895 /* Some amount of caching on the client is allowed */
1896 rc = ll_statfs_internal(ll_s2sbi(sb), &osfs, OBD_STATFS_SUM);
1900 statfs_unpack(sfs, &osfs);
1902 /* We need to downshift for all 32-bit kernels, because we can't
1903 * tell if the kernel is being called via sys_statfs64() or not.
1904 * Stop before overflowing f_bsize - in which case it is better
1905 * to just risk EOVERFLOW if caller is using old sys_statfs(). */
1906 if (sizeof(long) < 8) {
1907 while (osfs.os_blocks > ~0UL && sfs->f_bsize < 0x40000000) {
1910 osfs.os_blocks >>= 1;
1911 osfs.os_bfree >>= 1;
1912 osfs.os_bavail >>= 1;
1916 sfs->f_blocks = osfs.os_blocks;
1917 sfs->f_bfree = osfs.os_bfree;
1918 sfs->f_bavail = osfs.os_bavail;
1919 sfs->f_fsid.val[0] = (__u32)fsid;
1920 sfs->f_fsid.val[1] = (__u32)(fsid >> 32);
1924 void ll_inode_size_lock(struct inode *inode)
1926 struct ll_inode_info *lli;
1928 LASSERT(!S_ISDIR(inode->i_mode));
1930 lli = ll_i2info(inode);
1931 mutex_lock(&lli->lli_size_mutex);
1934 void ll_inode_size_unlock(struct inode *inode)
1936 struct ll_inode_info *lli;
1938 lli = ll_i2info(inode);
1939 mutex_unlock(&lli->lli_size_mutex);
1942 void ll_update_inode_flags(struct inode *inode, int ext_flags)
1944 inode->i_flags = ll_ext_to_inode_flags(ext_flags);
1945 if (ext_flags & LUSTRE_PROJINHERIT_FL)
1946 ll_file_set_flag(ll_i2info(inode), LLIF_PROJECT_INHERIT);
1948 ll_file_clear_flag(ll_i2info(inode), LLIF_PROJECT_INHERIT);
1951 int ll_update_inode(struct inode *inode, struct lustre_md *md)
1953 struct ll_inode_info *lli = ll_i2info(inode);
1954 struct mdt_body *body = md->body;
1955 struct ll_sb_info *sbi = ll_i2sbi(inode);
1958 if (body->mbo_valid & OBD_MD_FLEASIZE) {
1959 rc = cl_file_inode_init(inode, md);
1964 if (S_ISDIR(inode->i_mode)) {
1965 rc = ll_update_lsm_md(inode, md);
1970 #ifdef CONFIG_FS_POSIX_ACL
1971 if (body->mbo_valid & OBD_MD_FLACL) {
1972 spin_lock(&lli->lli_lock);
1973 if (lli->lli_posix_acl)
1974 posix_acl_release(lli->lli_posix_acl);
1975 lli->lli_posix_acl = md->posix_acl;
1976 spin_unlock(&lli->lli_lock);
1979 inode->i_ino = cl_fid_build_ino(&body->mbo_fid1,
1980 sbi->ll_flags & LL_SBI_32BIT_API);
1981 inode->i_generation = cl_fid_build_gen(&body->mbo_fid1);
1983 if (body->mbo_valid & OBD_MD_FLATIME) {
1984 if (body->mbo_atime > inode->i_atime.tv_sec)
1985 inode->i_atime.tv_sec = body->mbo_atime;
1986 lli->lli_atime = body->mbo_atime;
1989 if (body->mbo_valid & OBD_MD_FLMTIME) {
1990 if (body->mbo_mtime > inode->i_mtime.tv_sec) {
1992 "setting ino %lu mtime from %lld to %llu\n",
1993 inode->i_ino, (s64)inode->i_mtime.tv_sec,
1995 inode->i_mtime.tv_sec = body->mbo_mtime;
1997 lli->lli_mtime = body->mbo_mtime;
2000 if (body->mbo_valid & OBD_MD_FLCTIME) {
2001 if (body->mbo_ctime > inode->i_ctime.tv_sec)
2002 inode->i_ctime.tv_sec = body->mbo_ctime;
2003 lli->lli_ctime = body->mbo_ctime;
2006 /* Clear i_flags to remove S_NOSEC before permissions are updated */
2007 if (body->mbo_valid & OBD_MD_FLFLAGS)
2008 ll_update_inode_flags(inode, body->mbo_flags);
2009 if (body->mbo_valid & OBD_MD_FLMODE)
2010 inode->i_mode = (inode->i_mode & S_IFMT) |
2011 (body->mbo_mode & ~S_IFMT);
2013 if (body->mbo_valid & OBD_MD_FLTYPE)
2014 inode->i_mode = (inode->i_mode & ~S_IFMT) |
2015 (body->mbo_mode & S_IFMT);
2017 LASSERT(inode->i_mode != 0);
2018 if (S_ISREG(inode->i_mode))
2019 inode->i_blkbits = min(PTLRPC_MAX_BRW_BITS + 1,
2020 LL_MAX_BLKSIZE_BITS);
2022 inode->i_blkbits = inode->i_sb->s_blocksize_bits;
2024 if (body->mbo_valid & OBD_MD_FLUID)
2025 inode->i_uid = make_kuid(&init_user_ns, body->mbo_uid);
2026 if (body->mbo_valid & OBD_MD_FLGID)
2027 inode->i_gid = make_kgid(&init_user_ns, body->mbo_gid);
2028 if (body->mbo_valid & OBD_MD_FLPROJID)
2029 lli->lli_projid = body->mbo_projid;
2030 if (body->mbo_valid & OBD_MD_FLNLINK)
2031 set_nlink(inode, body->mbo_nlink);
2032 if (body->mbo_valid & OBD_MD_FLRDEV)
2033 inode->i_rdev = old_decode_dev(body->mbo_rdev);
2035 if (body->mbo_valid & OBD_MD_FLID) {
2036 /* FID shouldn't be changed! */
2037 if (fid_is_sane(&lli->lli_fid)) {
2038 LASSERTF(lu_fid_eq(&lli->lli_fid, &body->mbo_fid1),
2039 "Trying to change FID "DFID
2040 " to the "DFID", inode "DFID"(%p)\n",
2041 PFID(&lli->lli_fid), PFID(&body->mbo_fid1),
2042 PFID(ll_inode2fid(inode)), inode);
2044 lli->lli_fid = body->mbo_fid1;
2048 LASSERT(fid_seq(&lli->lli_fid) != 0);
2050 if (body->mbo_valid & OBD_MD_FLSIZE) {
2051 i_size_write(inode, body->mbo_size);
2053 CDEBUG(D_VFSTRACE, "inode="DFID", updating i_size %llu\n",
2054 PFID(ll_inode2fid(inode)),
2055 (unsigned long long)body->mbo_size);
2057 if (body->mbo_valid & OBD_MD_FLBLOCKS)
2058 inode->i_blocks = body->mbo_blocks;
2061 if (body->mbo_valid & OBD_MD_TSTATE) {
2062 /* Set LLIF_FILE_RESTORING if restore ongoing and
2063 * clear it when done to ensure to start again
2064 * glimpsing updated attrs
2066 if (body->mbo_t_state & MS_RESTORE)
2067 ll_file_set_flag(lli, LLIF_FILE_RESTORING);
2069 ll_file_clear_flag(lli, LLIF_FILE_RESTORING);
2075 int ll_read_inode2(struct inode *inode, void *opaque)
2077 struct lustre_md *md = opaque;
2078 struct ll_inode_info *lli = ll_i2info(inode);
2082 CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p)\n",
2083 PFID(&lli->lli_fid), inode);
2085 /* Core attributes from the MDS first. This is a new inode, and
2086 * the VFS doesn't zero times in the core inode so we have to do
2087 * it ourselves. They will be overwritten by either MDS or OST
2088 * attributes - we just need to make sure they aren't newer.
2090 inode->i_mtime.tv_sec = 0;
2091 inode->i_atime.tv_sec = 0;
2092 inode->i_ctime.tv_sec = 0;
2094 rc = ll_update_inode(inode, md);
2098 /* OIDEBUG(inode); */
2100 #ifdef HAVE_BACKING_DEV_INFO
2101 /* initializing backing dev info. */
2102 inode->i_mapping->backing_dev_info = &s2lsi(inode->i_sb)->lsi_bdi;
2104 if (S_ISREG(inode->i_mode)) {
2105 struct ll_sb_info *sbi = ll_i2sbi(inode);
2106 inode->i_op = &ll_file_inode_operations;
2107 inode->i_fop = sbi->ll_fop;
2108 inode->i_mapping->a_ops = (struct address_space_operations *)&ll_aops;
2110 } else if (S_ISDIR(inode->i_mode)) {
2111 inode->i_op = &ll_dir_inode_operations;
2112 inode->i_fop = &ll_dir_operations;
2114 } else if (S_ISLNK(inode->i_mode)) {
2115 inode->i_op = &ll_fast_symlink_inode_operations;
2118 inode->i_op = &ll_special_inode_operations;
2120 init_special_inode(inode, inode->i_mode,
2129 void ll_delete_inode(struct inode *inode)
2131 struct ll_inode_info *lli = ll_i2info(inode);
2132 struct address_space *mapping = &inode->i_data;
2133 unsigned long nrpages;
2136 if (S_ISREG(inode->i_mode) && lli->lli_clob != NULL)
2137 /* It is last chance to write out dirty pages,
2138 * otherwise we may lose data while umount */
2139 cl_sync_file_range(inode, 0, OBD_OBJECT_EOF, CL_FSYNC_LOCAL, 1);
2141 truncate_inode_pages_final(mapping);
2143 /* Workaround for LU-118: Note nrpages may not be totally updated when
2144 * truncate_inode_pages() returns, as there can be a page in the process
2145 * of deletion (inside __delete_from_page_cache()) in the specified
2146 * range. Thus mapping->nrpages can be non-zero when this function
2147 * returns even after truncation of the whole mapping. Only do this if
2148 * npages isn't already zero.
2150 nrpages = mapping->nrpages;
2152 xa_lock_irq(&mapping->i_pages);
2153 nrpages = mapping->nrpages;
2154 xa_unlock_irq(&mapping->i_pages);
2155 } /* Workaround end */
2157 LASSERTF(nrpages == 0, "%s: inode="DFID"(%p) nrpages=%lu, "
2158 "see https://jira.whamcloud.com/browse/LU-118\n",
2159 ll_i2sbi(inode)->ll_fsname,
2160 PFID(ll_inode2fid(inode)), inode, nrpages);
2162 #ifdef HAVE_SBOPS_EVICT_INODE
2163 ll_clear_inode(inode);
2170 int ll_iocontrol(struct inode *inode, struct file *file,
2171 unsigned int cmd, unsigned long arg)
2173 struct ll_sb_info *sbi = ll_i2sbi(inode);
2174 struct ptlrpc_request *req = NULL;
2179 case FS_IOC_GETFLAGS: {
2180 struct mdt_body *body;
2181 struct md_op_data *op_data;
2183 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL,
2184 0, 0, LUSTRE_OPC_ANY,
2186 if (IS_ERR(op_data))
2187 RETURN(PTR_ERR(op_data));
2189 op_data->op_valid = OBD_MD_FLFLAGS;
2190 rc = md_getattr(sbi->ll_md_exp, op_data, &req);
2191 ll_finish_md_op_data(op_data);
2193 CERROR("%s: failure inode "DFID": rc = %d\n",
2194 sbi->ll_md_exp->exp_obd->obd_name,
2195 PFID(ll_inode2fid(inode)), rc);
2199 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
2201 flags = body->mbo_flags;
2203 ptlrpc_req_finished(req);
2205 RETURN(put_user(flags, (int __user *)arg));
2207 case FS_IOC_SETFLAGS: {
2209 struct md_op_data *op_data;
2210 struct cl_object *obj;
2211 struct fsxattr fa = { 0 };
2213 if (get_user(flags, (int __user *)arg))
2216 fa.fsx_projid = ll_i2info(inode)->lli_projid;
2217 if (flags & LUSTRE_PROJINHERIT_FL)
2218 fa.fsx_xflags = FS_XFLAG_PROJINHERIT;
2220 rc = ll_ioctl_check_project(inode, &fa);
2224 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
2225 LUSTRE_OPC_ANY, NULL);
2226 if (IS_ERR(op_data))
2227 RETURN(PTR_ERR(op_data));
2229 op_data->op_attr_flags = flags;
2230 op_data->op_xvalid |= OP_XVALID_FLAGS;
2231 rc = md_setattr(sbi->ll_md_exp, op_data, NULL, 0, &req);
2232 ll_finish_md_op_data(op_data);
2233 ptlrpc_req_finished(req);
2237 ll_update_inode_flags(inode, flags);
2239 obj = ll_i2info(inode)->lli_clob;
2243 OBD_ALLOC_PTR(attr);
2247 rc = cl_setattr_ost(obj, attr, OP_XVALID_FLAGS, flags);
2259 int ll_flush_ctx(struct inode *inode)
2261 struct ll_sb_info *sbi = ll_i2sbi(inode);
2263 CDEBUG(D_SEC, "flush context for user %d\n",
2264 from_kuid(&init_user_ns, current_uid()));
2266 obd_set_info_async(NULL, sbi->ll_md_exp,
2267 sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
2269 obd_set_info_async(NULL, sbi->ll_dt_exp,
2270 sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
2275 /* umount -f client means force down, don't save state */
2276 void ll_umount_begin(struct super_block *sb)
2278 struct ll_sb_info *sbi = ll_s2sbi(sb);
2279 struct obd_device *obd;
2280 struct obd_ioctl_data *ioc_data;
2281 struct l_wait_info lwi;
2282 wait_queue_head_t waitq;
2285 CDEBUG(D_VFSTRACE, "VFS Op: superblock %p count %d active %d\n", sb,
2286 sb->s_count, atomic_read(&sb->s_active));
2288 obd = class_exp2obd(sbi->ll_md_exp);
2290 CERROR("Invalid MDC connection handle %#llx\n",
2291 sbi->ll_md_exp->exp_handle.h_cookie);
2297 obd = class_exp2obd(sbi->ll_dt_exp);
2299 CERROR("Invalid LOV connection handle %#llx\n",
2300 sbi->ll_dt_exp->exp_handle.h_cookie);
2306 OBD_ALLOC_PTR(ioc_data);
2308 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_md_exp,
2309 sizeof *ioc_data, ioc_data, NULL);
2311 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_dt_exp,
2312 sizeof *ioc_data, ioc_data, NULL);
2314 OBD_FREE_PTR(ioc_data);
2317 /* Really, we'd like to wait until there are no requests outstanding,
2318 * and then continue. For now, we just periodically checking for vfs
2319 * to decrement mnt_cnt and hope to finish it within 10sec.
2321 init_waitqueue_head(&waitq);
2322 lwi = LWI_TIMEOUT_INTERVAL(cfs_time_seconds(10),
2323 cfs_time_seconds(1), NULL, NULL);
2324 l_wait_event(waitq, may_umount(sbi->ll_mnt.mnt), &lwi);
2329 int ll_remount_fs(struct super_block *sb, int *flags, char *data)
2331 struct ll_sb_info *sbi = ll_s2sbi(sb);
2332 char *profilenm = get_profile_name(sb);
2336 if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY)) {
2337 read_only = *flags & MS_RDONLY;
2338 err = obd_set_info_async(NULL, sbi->ll_md_exp,
2339 sizeof(KEY_READ_ONLY),
2340 KEY_READ_ONLY, sizeof(read_only),
2343 LCONSOLE_WARN("Failed to remount %s %s (%d)\n",
2344 profilenm, read_only ?
2345 "read-only" : "read-write", err);
2350 sb->s_flags |= MS_RDONLY;
2352 sb->s_flags &= ~MS_RDONLY;
2354 if (sbi->ll_flags & LL_SBI_VERBOSE)
2355 LCONSOLE_WARN("Remounted %s %s\n", profilenm,
2356 read_only ? "read-only" : "read-write");
2362 * Cleanup the open handle that is cached on MDT-side.
2364 * For open case, the client side open handling thread may hit error
2365 * after the MDT grant the open. Under such case, the client should
2366 * send close RPC to the MDT as cleanup; otherwise, the open handle
2367 * on the MDT will be leaked there until the client umount or evicted.
2369 * In further, if someone unlinked the file, because the open handle
2370 * holds the reference on such file/object, then it will block the
2371 * subsequent threads that want to locate such object via FID.
2373 * \param[in] sb super block for this file-system
2374 * \param[in] open_req pointer to the original open request
2376 void ll_open_cleanup(struct super_block *sb, struct ptlrpc_request *open_req)
2378 struct mdt_body *body;
2379 struct md_op_data *op_data;
2380 struct ptlrpc_request *close_req = NULL;
2381 struct obd_export *exp = ll_s2sbi(sb)->ll_md_exp;
2384 body = req_capsule_server_get(&open_req->rq_pill, &RMF_MDT_BODY);
2385 OBD_ALLOC_PTR(op_data);
2386 if (op_data == NULL) {
2387 CWARN("%s: cannot allocate op_data to release open handle for "
2388 DFID"\n", ll_s2sbi(sb)->ll_fsname, PFID(&body->mbo_fid1));
2393 op_data->op_fid1 = body->mbo_fid1;
2394 op_data->op_open_handle = body->mbo_open_handle;
2395 op_data->op_mod_time = ktime_get_real_seconds();
2396 md_close(exp, op_data, NULL, &close_req);
2397 ptlrpc_req_finished(close_req);
2398 ll_finish_md_op_data(op_data);
2403 int ll_prep_inode(struct inode **inode, struct ptlrpc_request *req,
2404 struct super_block *sb, struct lookup_intent *it)
2406 struct ll_sb_info *sbi = NULL;
2407 struct lustre_md md = { NULL };
2411 LASSERT(*inode || sb);
2412 sbi = sb ? ll_s2sbi(sb) : ll_i2sbi(*inode);
2413 rc = md_get_lustre_md(sbi->ll_md_exp, req, sbi->ll_dt_exp,
2414 sbi->ll_md_exp, &md);
2419 rc = ll_update_inode(*inode, &md);
2423 LASSERT(sb != NULL);
2426 * At this point server returns to client's same fid as client
2427 * generated for creating. So using ->fid1 is okay here.
2429 if (!fid_is_sane(&md.body->mbo_fid1)) {
2430 CERROR("%s: Fid is insane "DFID"\n",
2432 PFID(&md.body->mbo_fid1));
2433 GOTO(out, rc = -EINVAL);
2436 *inode = ll_iget(sb, cl_fid_build_ino(&md.body->mbo_fid1,
2437 sbi->ll_flags & LL_SBI_32BIT_API),
2439 if (IS_ERR(*inode)) {
2440 #ifdef CONFIG_FS_POSIX_ACL
2442 posix_acl_release(md.posix_acl);
2443 md.posix_acl = NULL;
2446 rc = IS_ERR(*inode) ? PTR_ERR(*inode) : -ENOMEM;
2448 CERROR("new_inode -fatal: rc %d\n", rc);
2453 /* Handling piggyback layout lock.
2454 * Layout lock can be piggybacked by getattr and open request.
2455 * The lsm can be applied to inode only if it comes with a layout lock
2456 * otherwise correct layout may be overwritten, for example:
2457 * 1. proc1: mdt returns a lsm but not granting layout
2458 * 2. layout was changed by another client
2459 * 3. proc2: refresh layout and layout lock granted
2460 * 4. proc1: to apply a stale layout */
2461 if (it != NULL && it->it_lock_mode != 0) {
2462 struct lustre_handle lockh;
2463 struct ldlm_lock *lock;
2465 lockh.cookie = it->it_lock_handle;
2466 lock = ldlm_handle2lock(&lockh);
2467 LASSERT(lock != NULL);
2468 if (ldlm_has_layout(lock)) {
2469 struct cl_object_conf conf;
2471 memset(&conf, 0, sizeof(conf));
2472 conf.coc_opc = OBJECT_CONF_SET;
2473 conf.coc_inode = *inode;
2474 conf.coc_lock = lock;
2475 conf.u.coc_layout = md.layout;
2476 (void)ll_layout_conf(*inode, &conf);
2478 LDLM_LOCK_PUT(lock);
2484 /* cleanup will be done if necessary */
2485 md_free_lustre_md(sbi->ll_md_exp, &md);
2487 if (rc != 0 && it != NULL && it->it_op & IT_OPEN)
2488 ll_open_cleanup(sb != NULL ? sb : (*inode)->i_sb, req);
2493 int ll_obd_statfs(struct inode *inode, void __user *arg)
2495 struct ll_sb_info *sbi = NULL;
2496 struct obd_export *exp;
2498 struct obd_ioctl_data *data = NULL;
2502 if (!inode || !(sbi = ll_i2sbi(inode)))
2503 GOTO(out_statfs, rc = -EINVAL);
2505 rc = obd_ioctl_getdata(&buf, &len, arg);
2507 GOTO(out_statfs, rc);
2510 if (!data->ioc_inlbuf1 || !data->ioc_inlbuf2 ||
2511 !data->ioc_pbuf1 || !data->ioc_pbuf2)
2512 GOTO(out_statfs, rc = -EINVAL);
2514 if (data->ioc_inllen1 != sizeof(__u32) ||
2515 data->ioc_inllen2 != sizeof(__u32) ||
2516 data->ioc_plen1 != sizeof(struct obd_statfs) ||
2517 data->ioc_plen2 != sizeof(struct obd_uuid))
2518 GOTO(out_statfs, rc = -EINVAL);
2520 memcpy(&type, data->ioc_inlbuf1, sizeof(__u32));
2521 if (type & LL_STATFS_LMV)
2522 exp = sbi->ll_md_exp;
2523 else if (type & LL_STATFS_LOV)
2524 exp = sbi->ll_dt_exp;
2526 GOTO(out_statfs, rc = -ENODEV);
2528 rc = obd_iocontrol(IOC_OBD_STATFS, exp, len, buf, NULL);
2530 GOTO(out_statfs, rc);
2532 OBD_FREE_LARGE(buf, len);
2537 * this is normally called in ll_fini_md_op_data(), but sometimes it needs to
2538 * be called early to avoid deadlock.
2540 void ll_unlock_md_op_lsm(struct md_op_data *op_data)
2542 if (op_data->op_mea2_sem) {
2543 up_read(op_data->op_mea2_sem);
2544 op_data->op_mea2_sem = NULL;
2547 if (op_data->op_mea1_sem) {
2548 up_read(op_data->op_mea1_sem);
2549 op_data->op_mea1_sem = NULL;
2553 /* this function prepares md_op_data hint for passing it down to MD stack. */
2554 struct md_op_data *ll_prep_md_op_data(struct md_op_data *op_data,
2555 struct inode *i1, struct inode *i2,
2556 const char *name, size_t namelen,
2557 __u32 mode, __u32 opc, void *data)
2559 LASSERT(i1 != NULL);
2562 /* Do not reuse namelen for something else. */
2564 return ERR_PTR(-EINVAL);
2566 if (namelen > ll_i2sbi(i1)->ll_namelen)
2567 return ERR_PTR(-ENAMETOOLONG);
2569 if (!lu_name_is_valid_2(name, namelen))
2570 return ERR_PTR(-EINVAL);
2573 if (op_data == NULL)
2574 OBD_ALLOC_PTR(op_data);
2576 if (op_data == NULL)
2577 return ERR_PTR(-ENOMEM);
2579 ll_i2gids(op_data->op_suppgids, i1, i2);
2580 op_data->op_fid1 = *ll_inode2fid(i1);
2581 op_data->op_default_stripe_offset = -1;
2583 if (S_ISDIR(i1->i_mode)) {
2584 down_read(&ll_i2info(i1)->lli_lsm_sem);
2585 op_data->op_mea1_sem = &ll_i2info(i1)->lli_lsm_sem;
2586 op_data->op_mea1 = ll_i2info(i1)->lli_lsm_md;
2587 if (opc == LUSTRE_OPC_MKDIR)
2588 op_data->op_default_stripe_offset =
2589 ll_i2info(i1)->lli_def_stripe_offset;
2593 op_data->op_fid2 = *ll_inode2fid(i2);
2594 if (S_ISDIR(i2->i_mode)) {
2596 down_read(&ll_i2info(i2)->lli_lsm_sem);
2597 op_data->op_mea2_sem =
2598 &ll_i2info(i2)->lli_lsm_sem;
2600 op_data->op_mea2 = ll_i2info(i2)->lli_lsm_md;
2603 fid_zero(&op_data->op_fid2);
2606 if (ll_i2sbi(i1)->ll_flags & LL_SBI_64BIT_HASH)
2607 op_data->op_cli_flags |= CLI_HASH64;
2609 if (ll_need_32bit_api(ll_i2sbi(i1)))
2610 op_data->op_cli_flags |= CLI_API32;
2612 op_data->op_name = name;
2613 op_data->op_namelen = namelen;
2614 op_data->op_mode = mode;
2615 op_data->op_mod_time = ktime_get_real_seconds();
2616 op_data->op_fsuid = from_kuid(&init_user_ns, current_fsuid());
2617 op_data->op_fsgid = from_kgid(&init_user_ns, current_fsgid());
2618 op_data->op_cap = cfs_curproc_cap_pack();
2619 op_data->op_mds = 0;
2620 if ((opc == LUSTRE_OPC_CREATE) && (name != NULL) &&
2621 filename_is_volatile(name, namelen, &op_data->op_mds)) {
2622 op_data->op_bias |= MDS_CREATE_VOLATILE;
2624 op_data->op_data = data;
2629 void ll_finish_md_op_data(struct md_op_data *op_data)
2631 ll_unlock_md_op_lsm(op_data);
2632 security_release_secctx(op_data->op_file_secctx,
2633 op_data->op_file_secctx_size);
2634 OBD_FREE_PTR(op_data);
2637 #ifdef HAVE_SUPEROPS_USE_DENTRY
2638 int ll_show_options(struct seq_file *seq, struct dentry *dentry)
2640 int ll_show_options(struct seq_file *seq, struct vfsmount *vfs)
2643 struct ll_sb_info *sbi;
2645 #ifdef HAVE_SUPEROPS_USE_DENTRY
2646 LASSERT((seq != NULL) && (dentry != NULL));
2647 sbi = ll_s2sbi(dentry->d_sb);
2649 LASSERT((seq != NULL) && (vfs != NULL));
2650 sbi = ll_s2sbi(vfs->mnt_sb);
2653 if (sbi->ll_flags & LL_SBI_NOLCK)
2654 seq_puts(seq, ",nolock");
2656 if (sbi->ll_flags & LL_SBI_FLOCK)
2657 seq_puts(seq, ",flock");
2659 if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
2660 seq_puts(seq, ",localflock");
2662 if (sbi->ll_flags & LL_SBI_USER_XATTR)
2663 seq_puts(seq, ",user_xattr");
2665 if (sbi->ll_flags & LL_SBI_LAZYSTATFS)
2666 seq_puts(seq, ",lazystatfs");
2668 if (sbi->ll_flags & LL_SBI_USER_FID2PATH)
2669 seq_puts(seq, ",user_fid2path");
2671 if (sbi->ll_flags & LL_SBI_ALWAYS_PING)
2672 seq_puts(seq, ",always_ping");
2678 * Get obd name by cmd, and copy out to user space
2680 int ll_get_obd_name(struct inode *inode, unsigned int cmd, unsigned long arg)
2682 struct ll_sb_info *sbi = ll_i2sbi(inode);
2683 struct obd_device *obd;
2686 if (cmd == OBD_IOC_GETDTNAME)
2687 obd = class_exp2obd(sbi->ll_dt_exp);
2688 else if (cmd == OBD_IOC_GETMDNAME)
2689 obd = class_exp2obd(sbi->ll_md_exp);
2696 if (copy_to_user((void __user *)arg, obd->obd_name,
2697 strlen(obd->obd_name) + 1))
2703 static char* ll_d_path(struct dentry *dentry, char *buf, int bufsize)
2710 p.mnt = current->fs->root.mnt;
2712 path = d_path(&p, buf, bufsize);
2717 void ll_dirty_page_discard_warn(struct page *page, int ioret)
2719 char *buf, *path = NULL;
2720 struct dentry *dentry = NULL;
2721 struct inode *inode = page->mapping->host;
2723 /* this can be called inside spin lock so use GFP_ATOMIC. */
2724 buf = (char *)__get_free_page(GFP_ATOMIC);
2726 dentry = d_find_alias(page->mapping->host);
2728 path = ll_d_path(dentry, buf, PAGE_SIZE);
2732 "%s: dirty page discard: %s/fid: "DFID"/%s may get corrupted "
2733 "(rc %d)\n", ll_i2sbi(inode)->ll_fsname,
2734 s2lsi(page->mapping->host->i_sb)->lsi_lmd->lmd_dev,
2735 PFID(ll_inode2fid(inode)),
2736 (path && !IS_ERR(path)) ? path : "", ioret);
2742 free_page((unsigned long)buf);
2745 ssize_t ll_copy_user_md(const struct lov_user_md __user *md,
2746 struct lov_user_md **kbuf)
2748 struct lov_user_md lum;
2752 if (copy_from_user(&lum, md, sizeof(lum)))
2755 lum_size = ll_lov_user_md_size(&lum);
2759 OBD_ALLOC(*kbuf, lum_size);
2763 if (copy_from_user(*kbuf, md, lum_size) != 0) {
2764 OBD_FREE(*kbuf, lum_size);
2772 * Compute llite root squash state after a change of root squash
2773 * configuration setting or add/remove of a lnet nid
2775 void ll_compute_rootsquash_state(struct ll_sb_info *sbi)
2777 struct root_squash_info *squash = &sbi->ll_squash;
2780 struct lnet_process_id id;
2782 /* Update norootsquash flag */
2783 down_write(&squash->rsi_sem);
2784 if (list_empty(&squash->rsi_nosquash_nids))
2785 sbi->ll_flags &= ~LL_SBI_NOROOTSQUASH;
2787 /* Do not apply root squash as soon as one of our NIDs is
2788 * in the nosquash_nids list */
2791 while (LNetGetId(i++, &id) != -ENOENT) {
2792 if (LNET_NETTYP(LNET_NIDNET(id.nid)) == LOLND)
2794 if (cfs_match_nid(id.nid, &squash->rsi_nosquash_nids)) {
2800 sbi->ll_flags |= LL_SBI_NOROOTSQUASH;
2802 sbi->ll_flags &= ~LL_SBI_NOROOTSQUASH;
2804 up_write(&squash->rsi_sem);
2808 * Parse linkea content to extract information about a given hardlink
2810 * \param[in] ldata - Initialized linkea data
2811 * \param[in] linkno - Link identifier
2812 * \param[out] parent_fid - The entry's parent FID
2813 * \param[out] ln - Entry name destination buffer
2815 * \retval 0 on success
2816 * \retval Appropriate negative error code on failure
2818 static int ll_linkea_decode(struct linkea_data *ldata, unsigned int linkno,
2819 struct lu_fid *parent_fid, struct lu_name *ln)
2825 rc = linkea_init_with_rec(ldata);
2829 if (linkno >= ldata->ld_leh->leh_reccount)
2830 /* beyond last link */
2833 linkea_first_entry(ldata);
2834 for (idx = 0; ldata->ld_lee != NULL; idx++) {
2835 linkea_entry_unpack(ldata->ld_lee, &ldata->ld_reclen, ln,
2840 linkea_next_entry(ldata);
2850 * Get parent FID and name of an identified link. Operation is performed for
2851 * a given link number, letting the caller iterate over linkno to list one or
2852 * all links of an entry.
2854 * \param[in] file - File descriptor against which to perform the operation
2855 * \param[in,out] arg - User-filled structure containing the linkno to operate
2856 * on and the available size. It is eventually filled with
2857 * the requested information or left untouched on error
2859 * \retval - 0 on success
2860 * \retval - Appropriate negative error code on failure
2862 int ll_getparent(struct file *file, struct getparent __user *arg)
2864 struct inode *inode = file_inode(file);
2865 struct linkea_data *ldata;
2866 struct lu_buf buf = LU_BUF_NULL;
2868 struct lu_fid parent_fid;
2875 if (!cfs_capable(CFS_CAP_DAC_READ_SEARCH) &&
2876 !(ll_i2sbi(inode)->ll_flags & LL_SBI_USER_FID2PATH))
2879 if (get_user(name_size, &arg->gp_name_size))
2882 if (get_user(linkno, &arg->gp_linkno))
2885 if (name_size > PATH_MAX)
2888 OBD_ALLOC(ldata, sizeof(*ldata));
2892 rc = linkea_data_new(ldata, &buf);
2894 GOTO(ldata_free, rc);
2896 #ifdef HAVE_XATTR_HANDLER_FLAGS
2897 rc = ll_xattr_list(inode, XATTR_NAME_LINK, XATTR_TRUSTED_T, buf.lb_buf,
2898 buf.lb_len, OBD_MD_FLXATTR);
2900 rc = ll_getxattr(file_dentry(file), XATTR_NAME_LINK, buf.lb_buf,
2902 #endif /* HAVE_XATTR_HANDLER_FLAGS */
2906 rc = ll_linkea_decode(ldata, linkno, &parent_fid, &ln);
2910 if (ln.ln_namelen >= name_size)
2911 GOTO(lb_free, rc = -EOVERFLOW);
2913 if (copy_to_user(&arg->gp_fid, &parent_fid, sizeof(arg->gp_fid)))
2914 GOTO(lb_free, rc = -EFAULT);
2916 if (copy_to_user(&arg->gp_name, ln.ln_name, ln.ln_namelen))
2917 GOTO(lb_free, rc = -EFAULT);
2919 if (put_user('\0', arg->gp_name + ln.ln_namelen))
2920 GOTO(lb_free, rc = -EFAULT);
2925 OBD_FREE(ldata, sizeof(*ldata));