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);
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_FLR |
225 OBD_CONNECT2_LOCK_CONVERT |
226 OBD_CONNECT2_DIR_MIGRATE |
227 OBD_CONNECT2_SUM_STATFS |
228 OBD_CONNECT2_ARCHIVE_ID_ARRAY;
230 #ifdef HAVE_LRU_RESIZE_SUPPORT
231 if (sbi->ll_flags & LL_SBI_LRU_RESIZE)
232 data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
234 #ifdef CONFIG_FS_POSIX_ACL
235 data->ocd_connect_flags |= OBD_CONNECT_ACL | OBD_CONNECT_UMASK |
236 OBD_CONNECT_LARGE_ACL;
239 data->ocd_cksum_types = obd_cksum_types_supported_client();
241 if (OBD_FAIL_CHECK(OBD_FAIL_MDC_LIGHTWEIGHT))
242 /* flag mdc connection as lightweight, only used for test
243 * purpose, use with care */
244 data->ocd_connect_flags |= OBD_CONNECT_LIGHTWEIGHT;
246 data->ocd_ibits_known = MDS_INODELOCK_FULL;
247 data->ocd_version = LUSTRE_VERSION_CODE;
249 if (sb->s_flags & MS_RDONLY)
250 data->ocd_connect_flags |= OBD_CONNECT_RDONLY;
251 if (sbi->ll_flags & LL_SBI_USER_XATTR)
252 data->ocd_connect_flags |= OBD_CONNECT_XATTR;
255 /* Setting this indicates we correctly support S_NOSEC (See kernel
256 * commit 9e1f1de02c2275d7172e18dc4e7c2065777611bf)
258 sb->s_flags |= MS_NOSEC;
261 if (sbi->ll_flags & LL_SBI_FLOCK)
262 sbi->ll_fop = &ll_file_operations_flock;
263 else if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
264 sbi->ll_fop = &ll_file_operations;
266 sbi->ll_fop = &ll_file_operations_noflock;
268 /* always ping even if server suppress_pings */
269 if (sbi->ll_flags & LL_SBI_ALWAYS_PING)
270 data->ocd_connect_flags &= ~OBD_CONNECT_PINGLESS;
272 obd_connect_set_secctx(data);
274 data->ocd_brw_size = MD_MAX_BRW_SIZE;
276 err = obd_connect(NULL, &sbi->ll_md_exp, sbi->ll_md_obd,
277 &sbi->ll_sb_uuid, data, NULL);
279 LCONSOLE_ERROR_MSG(0x14f, "An MDT (md %s) is performing "
280 "recovery, of which this client is not a "
281 "part. Please wait for recovery to complete,"
282 " abort, or time out.\n", md);
285 CERROR("cannot connect to %s: rc = %d\n", md, err);
289 sbi->ll_md_exp->exp_connect_data = *data;
291 err = obd_fid_init(sbi->ll_md_exp->exp_obd, sbi->ll_md_exp,
292 LUSTRE_SEQ_METADATA);
294 CERROR("%s: Can't init metadata layer FID infrastructure, "
295 "rc = %d\n", sbi->ll_md_exp->exp_obd->obd_name, err);
299 /* For mount, we only need fs info from MDT0, and also in DNE, it
300 * can make sure the client can be mounted as long as MDT0 is
302 err = obd_statfs(NULL, sbi->ll_md_exp, osfs,
303 ktime_get_seconds() - OBD_STATFS_CACHE_SECONDS,
304 OBD_STATFS_FOR_MDT0);
306 GOTO(out_md_fid, err);
308 /* This needs to be after statfs to ensure connect has finished.
309 * Note that "data" does NOT contain the valid connect reply.
310 * If connecting to a 1.8 server there will be no LMV device, so
311 * we can access the MDC export directly and exp_connect_flags will
312 * be non-zero, but if accessing an upgraded 2.1 server it will
313 * have the correct flags filled in.
314 * XXX: fill in the LMV exp_connect_flags from MDC(s). */
315 valid = exp_connect_flags(sbi->ll_md_exp) & CLIENT_CONNECT_MDT_REQD;
316 if (exp_connect_flags(sbi->ll_md_exp) != 0 &&
317 valid != CLIENT_CONNECT_MDT_REQD) {
320 OBD_ALLOC_WAIT(buf, PAGE_SIZE);
321 obd_connect_flags2str(buf, PAGE_SIZE,
322 valid ^ CLIENT_CONNECT_MDT_REQD, 0, ",");
323 LCONSOLE_ERROR_MSG(0x170, "Server %s does not support "
324 "feature(s) needed for correct operation "
325 "of this client (%s). Please upgrade "
326 "server or downgrade client.\n",
327 sbi->ll_md_exp->exp_obd->obd_name, buf);
328 OBD_FREE(buf, PAGE_SIZE);
329 GOTO(out_md_fid, err = -EPROTO);
332 size = sizeof(*data);
333 err = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_CONN_DATA),
334 KEY_CONN_DATA, &size, data);
336 CERROR("%s: Get connect data failed: rc = %d\n",
337 sbi->ll_md_exp->exp_obd->obd_name, err);
338 GOTO(out_md_fid, err);
341 LASSERT(osfs->os_bsize);
342 sb->s_blocksize = osfs->os_bsize;
343 sb->s_blocksize_bits = log2(osfs->os_bsize);
344 sb->s_magic = LL_SUPER_MAGIC;
345 sb->s_maxbytes = MAX_LFS_FILESIZE;
346 sbi->ll_namelen = osfs->os_namelen;
347 sbi->ll_mnt.mnt = current->fs->root.mnt;
349 if ((sbi->ll_flags & LL_SBI_USER_XATTR) &&
350 !(data->ocd_connect_flags & OBD_CONNECT_XATTR)) {
351 LCONSOLE_INFO("Disabling user_xattr feature because "
352 "it is not supported on the server\n");
353 sbi->ll_flags &= ~LL_SBI_USER_XATTR;
356 if (data->ocd_connect_flags & OBD_CONNECT_ACL) {
358 sb->s_flags |= MS_POSIXACL;
360 sbi->ll_flags |= LL_SBI_ACL;
362 LCONSOLE_INFO("client wants to enable acl, but mdt not!\n");
364 sb->s_flags &= ~MS_POSIXACL;
366 sbi->ll_flags &= ~LL_SBI_ACL;
369 if (data->ocd_connect_flags & OBD_CONNECT_64BITHASH)
370 sbi->ll_flags |= LL_SBI_64BIT_HASH;
372 if (data->ocd_connect_flags & OBD_CONNECT_LAYOUTLOCK)
373 sbi->ll_flags |= LL_SBI_LAYOUT_LOCK;
375 if (obd_connect_has_secctx(data))
376 sbi->ll_flags |= LL_SBI_FILE_SECCTX;
378 if (data->ocd_ibits_known & MDS_INODELOCK_XATTR) {
379 if (!(data->ocd_connect_flags & OBD_CONNECT_MAX_EASIZE)) {
380 LCONSOLE_INFO("%s: disabling xattr cache due to "
381 "unknown maximum xattr size.\n", dt);
382 } else if (!sbi->ll_xattr_cache_set) {
383 /* If xattr_cache is already set (no matter 0 or 1)
384 * during processing llog, it won't be enabled here. */
385 sbi->ll_flags |= LL_SBI_XATTR_CACHE;
386 sbi->ll_xattr_cache_enabled = 1;
390 sbi->ll_dt_obd = class_name2obd(dt);
391 if (!sbi->ll_dt_obd) {
392 CERROR("DT %s: not setup or attached\n", dt);
393 GOTO(out_md_fid, err = -ENODEV);
396 /* pass client page size via ocd_grant_blkbits, the server should report
397 * back its backend blocksize for grant calculation purpose */
398 data->ocd_grant_blkbits = PAGE_SHIFT;
400 /* indicate OST features supported by this client */
401 data->ocd_connect_flags = OBD_CONNECT_GRANT | OBD_CONNECT_VERSION |
402 OBD_CONNECT_REQPORTAL | OBD_CONNECT_BRW_SIZE |
403 OBD_CONNECT_CANCELSET | OBD_CONNECT_FID |
404 OBD_CONNECT_SRVLOCK | OBD_CONNECT_TRUNCLOCK|
405 OBD_CONNECT_AT | OBD_CONNECT_OSS_CAPA |
406 OBD_CONNECT_VBR | OBD_CONNECT_FULL20 |
407 OBD_CONNECT_64BITHASH | OBD_CONNECT_MAXBYTES |
408 OBD_CONNECT_EINPROGRESS |
409 OBD_CONNECT_JOBSTATS | OBD_CONNECT_LVB_TYPE |
410 OBD_CONNECT_LAYOUTLOCK |
411 OBD_CONNECT_PINGLESS | OBD_CONNECT_LFSCK |
412 OBD_CONNECT_BULK_MBITS | OBD_CONNECT_SHORTIO |
413 OBD_CONNECT_FLAGS2 | OBD_CONNECT_GRANT_SHRINK;
415 /* The client currently advertises support for OBD_CONNECT_LOCKAHEAD_OLD so it
416 * can interoperate with an older version of lockahead which was released prior
417 * to landing in master. This support will be dropped when 2.13 development
418 * starts. At the point, we should not just drop the connect flag (below), we
419 * should also remove the support in the code.
421 * Removing it means a few things:
422 * 1. Remove this section here
423 * 2. Remove CEF_NONBLOCK in ll_file_lockahead()
424 * 3. Remove function exp_connect_lockahead_old
425 * 4. Remove LDLM_FL_LOCKAHEAD_OLD_RESERVED in lustre_dlm_flags.h
427 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 12, 50, 0)
428 data->ocd_connect_flags |= OBD_CONNECT_LOCKAHEAD_OLD;
431 data->ocd_connect_flags2 = OBD_CONNECT2_LOCKAHEAD;
433 if (!OBD_FAIL_CHECK(OBD_FAIL_OSC_CONNECT_GRANT_PARAM))
434 data->ocd_connect_flags |= OBD_CONNECT_GRANT_PARAM;
436 /* OBD_CONNECT_CKSUM should always be set, even if checksums are
437 * disabled by default, because it can still be enabled on the
438 * fly via /sys. As a consequence, we still need to come to an
439 * agreement on the supported algorithms at connect time
441 data->ocd_connect_flags |= OBD_CONNECT_CKSUM;
443 if (OBD_FAIL_CHECK(OBD_FAIL_OSC_CKSUM_ADLER_ONLY))
444 data->ocd_cksum_types = OBD_CKSUM_ADLER;
446 data->ocd_cksum_types = obd_cksum_types_supported_client();
448 #ifdef HAVE_LRU_RESIZE_SUPPORT
449 data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
451 /* always ping even if server suppress_pings */
452 if (sbi->ll_flags & LL_SBI_ALWAYS_PING)
453 data->ocd_connect_flags &= ~OBD_CONNECT_PINGLESS;
455 CDEBUG(D_RPCTRACE, "ocd_connect_flags: %#llx ocd_version: %d "
456 "ocd_grant: %d\n", data->ocd_connect_flags,
457 data->ocd_version, data->ocd_grant);
459 sbi->ll_dt_obd->obd_upcall.onu_owner = &sbi->ll_lco;
460 sbi->ll_dt_obd->obd_upcall.onu_upcall = cl_ocd_update;
462 data->ocd_brw_size = DT_MAX_BRW_SIZE;
464 err = obd_connect(NULL, &sbi->ll_dt_exp, sbi->ll_dt_obd,
465 &sbi->ll_sb_uuid, data, NULL);
467 LCONSOLE_ERROR_MSG(0x150, "An OST (dt %s) is performing "
468 "recovery, of which this client is not a "
469 "part. Please wait for recovery to "
470 "complete, abort, or time out.\n", dt);
473 CERROR("%s: Cannot connect to %s: rc = %d\n",
474 sbi->ll_dt_exp->exp_obd->obd_name, dt, err);
478 sbi->ll_dt_exp->exp_connect_data = *data;
480 /* Don't change value if it was specified in the config log */
481 if (sbi->ll_ra_info.ra_max_read_ahead_whole_pages == -1)
482 sbi->ll_ra_info.ra_max_read_ahead_whole_pages =
483 max_t(unsigned long, SBI_DEFAULT_READAHEAD_WHOLE_MAX,
484 (data->ocd_brw_size >> PAGE_SHIFT));
486 err = obd_fid_init(sbi->ll_dt_exp->exp_obd, sbi->ll_dt_exp,
487 LUSTRE_SEQ_METADATA);
489 CERROR("%s: Can't init data layer FID infrastructure, "
490 "rc = %d\n", sbi->ll_dt_exp->exp_obd->obd_name, err);
494 mutex_lock(&sbi->ll_lco.lco_lock);
495 sbi->ll_lco.lco_flags = data->ocd_connect_flags;
496 sbi->ll_lco.lco_md_exp = sbi->ll_md_exp;
497 sbi->ll_lco.lco_dt_exp = sbi->ll_dt_exp;
498 mutex_unlock(&sbi->ll_lco.lco_lock);
500 fid_zero(&sbi->ll_root_fid);
501 err = md_get_root(sbi->ll_md_exp, get_mount_fileset(sb),
504 CERROR("cannot mds_connect: rc = %d\n", err);
505 GOTO(out_lock_cn_cb, err);
507 if (!fid_is_sane(&sbi->ll_root_fid)) {
508 CERROR("%s: Invalid root fid "DFID" during mount\n",
509 sbi->ll_md_exp->exp_obd->obd_name,
510 PFID(&sbi->ll_root_fid));
511 GOTO(out_lock_cn_cb, err = -EINVAL);
513 CDEBUG(D_SUPER, "rootfid "DFID"\n", PFID(&sbi->ll_root_fid));
515 sb->s_op = &lustre_super_operations;
516 #ifdef HAVE_XATTR_HANDLER_FLAGS
517 sb->s_xattr = ll_xattr_handlers;
519 #if THREAD_SIZE >= 8192 /*b=17630*/
520 sb->s_export_op = &lustre_export_operations;
524 * XXX: move this to after cbd setup? */
525 valid = OBD_MD_FLGETATTR | OBD_MD_FLBLOCKS | OBD_MD_FLMODEASIZE;
526 if (sbi->ll_flags & LL_SBI_ACL)
527 valid |= OBD_MD_FLACL;
529 OBD_ALLOC_PTR(op_data);
531 GOTO(out_lock_cn_cb, err = -ENOMEM);
533 op_data->op_fid1 = sbi->ll_root_fid;
534 op_data->op_mode = 0;
535 op_data->op_valid = valid;
537 err = md_getattr(sbi->ll_md_exp, op_data, &request);
539 OBD_FREE_PTR(op_data);
541 CERROR("%s: md_getattr failed for root: rc = %d\n",
542 sbi->ll_md_exp->exp_obd->obd_name, err);
543 GOTO(out_lock_cn_cb, err);
546 err = md_get_lustre_md(sbi->ll_md_exp, request, sbi->ll_dt_exp,
547 sbi->ll_md_exp, &lmd);
549 CERROR("failed to understand root inode md: rc = %d\n", err);
550 ptlrpc_req_finished(request);
551 GOTO(out_lock_cn_cb, err);
554 LASSERT(fid_is_sane(&sbi->ll_root_fid));
555 root = ll_iget(sb, cl_fid_build_ino(&sbi->ll_root_fid,
556 sbi->ll_flags & LL_SBI_32BIT_API),
558 md_free_lustre_md(sbi->ll_md_exp, &lmd);
559 ptlrpc_req_finished(request);
562 #ifdef CONFIG_FS_POSIX_ACL
564 posix_acl_release(lmd.posix_acl);
565 lmd.posix_acl = NULL;
568 err = IS_ERR(root) ? PTR_ERR(root) : -EBADF;
570 CERROR("lustre_lite: bad iget4 for root\n");
574 checksum = sbi->ll_flags & LL_SBI_CHECKSUM;
575 if (sbi->ll_checksum_set) {
576 err = obd_set_info_async(NULL, sbi->ll_dt_exp,
577 sizeof(KEY_CHECKSUM), KEY_CHECKSUM,
578 sizeof(checksum), &checksum, NULL);
580 CERROR("%s: Set checksum failed: rc = %d\n",
581 sbi->ll_dt_exp->exp_obd->obd_name, err);
587 err = obd_set_info_async(NULL, sbi->ll_dt_exp, sizeof(KEY_CACHE_SET),
588 KEY_CACHE_SET, sizeof(*sbi->ll_cache),
589 sbi->ll_cache, NULL);
591 CERROR("%s: Set cache_set failed: rc = %d\n",
592 sbi->ll_dt_exp->exp_obd->obd_name, err);
596 sb->s_root = d_make_root(root);
597 if (sb->s_root == NULL) {
598 CERROR("%s: can't make root dentry\n",
599 ll_get_fsname(sb, NULL, 0));
600 GOTO(out_root, err = -ENOMEM);
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, ll_get_fsname(sb, NULL, 0), 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, ll_get_fsname(sb, NULL, 0), 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);
682 rc = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_MAX_EASIZE),
683 KEY_MAX_EASIZE, &size, lmmsize);
685 CERROR("Get max mdsize error rc %d\n", rc);
691 * Get the value of the default_easize parameter.
693 * \see client_obd::cl_default_mds_easize
695 * \param[in] sbi superblock info for this filesystem
696 * \param[out] lmmsize pointer to storage location for value
698 * \retval 0 on success
699 * \retval negative negated errno on failure
701 int ll_get_default_mdsize(struct ll_sb_info *sbi, int *lmmsize)
706 rc = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_DEFAULT_EASIZE),
707 KEY_DEFAULT_EASIZE, &size, lmmsize);
709 CERROR("Get default mdsize error rc %d\n", rc);
715 * Set the default_easize parameter to the given value.
717 * \see client_obd::cl_default_mds_easize
719 * \param[in] sbi superblock info for this filesystem
720 * \param[in] lmmsize the size to set
722 * \retval 0 on success
723 * \retval negative negated errno on failure
725 int ll_set_default_mdsize(struct ll_sb_info *sbi, int lmmsize)
729 if (lmmsize < sizeof(struct lov_mds_md) ||
730 lmmsize > OBD_MAX_DEFAULT_EA_SIZE)
733 rc = obd_set_info_async(NULL, sbi->ll_md_exp,
734 sizeof(KEY_DEFAULT_EASIZE), KEY_DEFAULT_EASIZE,
735 sizeof(int), &lmmsize, NULL);
740 static void client_common_put_super(struct super_block *sb)
742 struct ll_sb_info *sbi = ll_s2sbi(sb);
747 obd_fid_fini(sbi->ll_dt_exp->exp_obd);
748 obd_disconnect(sbi->ll_dt_exp);
749 sbi->ll_dt_exp = NULL;
751 ll_debugfs_unregister_super(sb);
753 obd_fid_fini(sbi->ll_md_exp->exp_obd);
754 obd_disconnect(sbi->ll_md_exp);
755 sbi->ll_md_exp = NULL;
760 void ll_kill_super(struct super_block *sb)
762 struct ll_sb_info *sbi;
766 if (!(sb->s_flags & MS_ACTIVE))
770 /* we need restore s_dev from changed for clustred NFS before put_super
771 * because new kernels have cached s_dev and change sb->s_dev in
772 * put_super not affected real removing devices */
774 sb->s_dev = sbi->ll_sdev_orig;
775 sbi->ll_umounting = 1;
777 /* wait running statahead threads to quit */
778 while (atomic_read(&sbi->ll_sa_running) > 0) {
779 set_current_state(TASK_UNINTERRUPTIBLE);
780 schedule_timeout(msecs_to_jiffies(MSEC_PER_SEC >> 3));
787 static inline int ll_set_opt(const char *opt, char *data, int fl)
789 if (strncmp(opt, data, strlen(opt)) != 0)
795 /* non-client-specific mount options are parsed in lmd_parse */
796 static int ll_options(char *options, struct ll_sb_info *sbi)
799 char *s1 = options, *s2;
800 int *flags = &sbi->ll_flags;
806 CDEBUG(D_CONFIG, "Parsing opts %s\n", options);
809 CDEBUG(D_SUPER, "next opt=%s\n", s1);
810 tmp = ll_set_opt("nolock", s1, LL_SBI_NOLCK);
815 tmp = ll_set_opt("flock", s1, LL_SBI_FLOCK);
820 tmp = ll_set_opt("localflock", s1, LL_SBI_LOCALFLOCK);
825 tmp = ll_set_opt("noflock", s1, LL_SBI_FLOCK|LL_SBI_LOCALFLOCK);
830 tmp = ll_set_opt("user_xattr", s1, LL_SBI_USER_XATTR);
835 tmp = ll_set_opt("nouser_xattr", s1, LL_SBI_USER_XATTR);
840 tmp = ll_set_opt("context", s1, 1);
843 tmp = ll_set_opt("fscontext", s1, 1);
846 tmp = ll_set_opt("defcontext", s1, 1);
849 tmp = ll_set_opt("rootcontext", s1, 1);
852 tmp = ll_set_opt("user_fid2path", s1, LL_SBI_USER_FID2PATH);
857 tmp = ll_set_opt("nouser_fid2path", s1, LL_SBI_USER_FID2PATH);
863 tmp = ll_set_opt("checksum", s1, LL_SBI_CHECKSUM);
866 sbi->ll_checksum_set = 1;
869 tmp = ll_set_opt("nochecksum", s1, LL_SBI_CHECKSUM);
872 sbi->ll_checksum_set = 1;
875 tmp = ll_set_opt("lruresize", s1, LL_SBI_LRU_RESIZE);
880 tmp = ll_set_opt("nolruresize", s1, LL_SBI_LRU_RESIZE);
885 tmp = ll_set_opt("lazystatfs", s1, LL_SBI_LAZYSTATFS);
890 tmp = ll_set_opt("nolazystatfs", s1, LL_SBI_LAZYSTATFS);
895 tmp = ll_set_opt("32bitapi", s1, LL_SBI_32BIT_API);
900 tmp = ll_set_opt("verbose", s1, LL_SBI_VERBOSE);
905 tmp = ll_set_opt("noverbose", s1, LL_SBI_VERBOSE);
910 tmp = ll_set_opt("always_ping", s1, LL_SBI_ALWAYS_PING);
915 LCONSOLE_ERROR_MSG(0x152, "Unknown option '%s', won't mount.\n",
921 s2 = strchr(s1, ',');
929 void ll_lli_init(struct ll_inode_info *lli)
931 lli->lli_inode_magic = LLI_INODE_MAGIC;
933 spin_lock_init(&lli->lli_lock);
934 lli->lli_posix_acl = NULL;
935 /* Do not set lli_fid, it has been initialized already. */
936 fid_zero(&lli->lli_pfid);
937 lli->lli_mds_read_och = NULL;
938 lli->lli_mds_write_och = NULL;
939 lli->lli_mds_exec_och = NULL;
940 lli->lli_open_fd_read_count = 0;
941 lli->lli_open_fd_write_count = 0;
942 lli->lli_open_fd_exec_count = 0;
943 mutex_init(&lli->lli_och_mutex);
944 spin_lock_init(&lli->lli_agl_lock);
945 spin_lock_init(&lli->lli_layout_lock);
946 ll_layout_version_set(lli, CL_LAYOUT_GEN_NONE);
947 lli->lli_clob = NULL;
949 init_rwsem(&lli->lli_xattrs_list_rwsem);
950 mutex_init(&lli->lli_xattrs_enq_lock);
952 LASSERT(lli->lli_vfs_inode.i_mode != 0);
953 if (S_ISDIR(lli->lli_vfs_inode.i_mode)) {
954 mutex_init(&lli->lli_readdir_mutex);
955 lli->lli_opendir_key = NULL;
957 spin_lock_init(&lli->lli_sa_lock);
958 lli->lli_opendir_pid = 0;
959 lli->lli_sa_enabled = 0;
960 lli->lli_def_stripe_offset = -1;
961 init_rwsem(&lli->lli_lsm_sem);
963 mutex_init(&lli->lli_size_mutex);
964 lli->lli_symlink_name = NULL;
965 init_rwsem(&lli->lli_trunc_sem);
966 range_lock_tree_init(&lli->lli_write_tree);
967 init_rwsem(&lli->lli_glimpse_sem);
968 lli->lli_glimpse_time = ktime_set(0, 0);
969 INIT_LIST_HEAD(&lli->lli_agl_list);
970 lli->lli_agl_index = 0;
971 lli->lli_async_rc = 0;
973 mutex_init(&lli->lli_layout_mutex);
974 memset(lli->lli_jobid, 0, sizeof(lli->lli_jobid));
977 #define MAX_STRING_SIZE 128
979 #ifndef HAVE_SUPER_SETUP_BDI_NAME
981 #define LSI_BDI_INITIALIZED 0x00400000
983 #ifndef HAVE_BDI_CAP_MAP_COPY
984 # define BDI_CAP_MAP_COPY 0
987 static int super_setup_bdi_name(struct super_block *sb, char *fmt, ...)
989 struct lustre_sb_info *lsi = s2lsi(sb);
990 char buf[MAX_STRING_SIZE];
994 err = bdi_init(&lsi->lsi_bdi);
998 lsi->lsi_flags |= LSI_BDI_INITIALIZED;
999 lsi->lsi_bdi.capabilities = BDI_CAP_MAP_COPY;
1000 lsi->lsi_bdi.name = "lustre";
1001 va_start(args, fmt);
1002 vsnprintf(buf, MAX_STRING_SIZE, fmt, args);
1004 err = bdi_register(&lsi->lsi_bdi, NULL, "%s", buf);
1007 sb->s_bdi = &lsi->lsi_bdi;
1011 #endif /* !HAVE_SUPER_SETUP_BDI_NAME */
1013 int ll_fill_super(struct super_block *sb, struct vfsmount *mnt)
1015 struct lustre_profile *lprof = NULL;
1016 struct lustre_sb_info *lsi = s2lsi(sb);
1017 struct ll_sb_info *sbi = NULL;
1018 char *dt = NULL, *md = NULL;
1019 char *profilenm = get_profile_name(sb);
1020 struct config_llog_instance *cfg;
1021 /* %p for void* in printf needs 16+2 characters: 0xffffffffffffffff */
1022 const int instlen = 16 + 2;
1023 unsigned long cfg_instance = ll_get_cfg_instance(sb);
1024 char name[MAX_STRING_SIZE];
1032 /* for ASLR, to map between cfg_instance and hashed ptr */
1033 CDEBUG(D_VFSTRACE, "VFS Op: cfg_instance %s-%016lx (sb %p)\n",
1034 profilenm, cfg_instance, sb);
1036 try_module_get(THIS_MODULE);
1040 GOTO(out_free_cfg, err = -ENOMEM);
1042 /* client additional sb info */
1043 lsi->lsi_llsbi = sbi = ll_init_sbi();
1045 GOTO(out_free_cfg, err = -ENOMEM);
1047 err = ll_options(lsi->lsi_lmd->lmd_opts, sbi);
1049 GOTO(out_free_cfg, err);
1051 err = super_setup_bdi_name(sb, "lustre-%016lx", cfg_instance);
1053 GOTO(out_free_cfg, err);
1055 #ifndef HAVE_DCACHE_LOCK
1056 /* kernel >= 2.6.38 store dentry operations in sb->s_d_op. */
1057 sb->s_d_op = &ll_d_ops;
1060 len = strlen(profilenm);
1061 ptr = strrchr(profilenm, '-');
1062 if (ptr && (strcmp(ptr, "-client") == 0))
1066 snprintf(name, MAX_STRING_SIZE, "%.*s-%016lx", len,
1067 profilenm, cfg_instance);
1069 /* Call ll_debugfs_register_super() before lustre_process_log()
1070 * so that "llite.*.*" params can be processed correctly.
1072 err = ll_debugfs_register_super(sb, name);
1074 CERROR("%s: could not register mountpoint in llite: rc = %d\n",
1075 ll_get_fsname(sb, NULL, 0), err);
1079 /* The cfg_instance is a value unique to this super, in case some
1080 * joker tries to mount the same fs at two mount points.
1082 cfg->cfg_instance = cfg_instance;
1083 cfg->cfg_uuid = lsi->lsi_llsbi->ll_sb_uuid;
1084 cfg->cfg_callback = class_config_llog_handler;
1085 cfg->cfg_sub_clds = CONFIG_SUB_CLIENT;
1086 /* set up client obds */
1087 err = lustre_process_log(sb, profilenm, cfg);
1089 GOTO(out_debugfs, err);
1091 /* Profile set with LCFG_MOUNTOPT so we can find our mdc and osc obds */
1092 lprof = class_get_profile(profilenm);
1093 if (lprof == NULL) {
1094 LCONSOLE_ERROR_MSG(0x156, "The client profile '%s' could not be"
1095 " read from the MGS. Does that filesystem "
1096 "exist?\n", profilenm);
1097 GOTO(out_debugfs, err = -EINVAL);
1099 CDEBUG(D_CONFIG, "Found profile %s: mdc=%s osc=%s\n", profilenm,
1100 lprof->lp_md, lprof->lp_dt);
1102 dt_len = strlen(lprof->lp_dt) + instlen + 2;
1103 OBD_ALLOC(dt, dt_len);
1105 GOTO(out_profile, err = -ENOMEM);
1106 snprintf(dt, dt_len - 1, "%s-%016lx", lprof->lp_dt, cfg_instance);
1108 md_len = strlen(lprof->lp_md) + instlen + 2;
1109 OBD_ALLOC(md, md_len);
1111 GOTO(out_free_dt, err = -ENOMEM);
1112 snprintf(md, md_len - 1, "%s-%016lx", lprof->lp_md, cfg_instance);
1114 /* connections, registrations, sb setup */
1115 err = client_common_fill_super(sb, md, dt, mnt);
1117 GOTO(out_free_md, err);
1119 sbi->ll_client_common_fill_super_succeeded = 1;
1123 OBD_FREE(md, md_len);
1126 OBD_FREE(dt, dt_len);
1129 class_put_profile(lprof);
1132 ll_debugfs_unregister_super(sb);
1139 else if (sbi->ll_flags & LL_SBI_VERBOSE)
1140 LCONSOLE_WARN("Mounted %s\n", profilenm);
1142 } /* ll_fill_super */
1144 void ll_put_super(struct super_block *sb)
1146 struct config_llog_instance cfg, params_cfg;
1147 struct obd_device *obd;
1148 struct lustre_sb_info *lsi = s2lsi(sb);
1149 struct ll_sb_info *sbi = ll_s2sbi(sb);
1150 char *profilenm = get_profile_name(sb);
1151 unsigned long cfg_instance = ll_get_cfg_instance(sb);
1153 int next, force = 1, rc = 0;
1157 GOTO(out_no_sbi, 0);
1159 /* Should replace instance_id with something better for ASLR */
1160 CDEBUG(D_VFSTRACE, "VFS Op: cfg_instance %s-%016lx (sb %p)\n",
1161 profilenm, cfg_instance, sb);
1163 cfg.cfg_instance = cfg_instance;
1164 lustre_end_log(sb, profilenm, &cfg);
1166 params_cfg.cfg_instance = cfg_instance;
1167 lustre_end_log(sb, PARAMS_FILENAME, ¶ms_cfg);
1169 if (sbi->ll_md_exp) {
1170 obd = class_exp2obd(sbi->ll_md_exp);
1172 force = obd->obd_force;
1175 /* Wait for unstable pages to be committed to stable storage */
1177 struct l_wait_info lwi = LWI_INTR(LWI_ON_SIGNAL_NOOP, NULL);
1178 rc = l_wait_event(sbi->ll_cache->ccc_unstable_waitq,
1179 atomic_long_read(&sbi->ll_cache->ccc_unstable_nr) == 0,
1183 ccc_count = atomic_long_read(&sbi->ll_cache->ccc_unstable_nr);
1184 if (force == 0 && rc != -EINTR)
1185 LASSERTF(ccc_count == 0, "count: %li\n", ccc_count);
1187 /* We need to set force before the lov_disconnect in
1188 lustre_common_put_super, since l_d cleans up osc's as well. */
1191 while ((obd = class_devices_in_group(&sbi->ll_sb_uuid,
1193 obd->obd_force = force;
1197 if (sbi->ll_client_common_fill_super_succeeded) {
1198 /* Only if client_common_fill_super succeeded */
1199 client_common_put_super(sb);
1203 while ((obd = class_devices_in_group(&sbi->ll_sb_uuid, &next)) !=NULL) {
1204 class_manual_cleanup(obd);
1207 if (sbi->ll_flags & LL_SBI_VERBOSE)
1208 LCONSOLE_WARN("Unmounted %s\n", profilenm ? profilenm : "");
1211 class_del_profile(profilenm);
1213 #ifndef HAVE_SUPER_SETUP_BDI_NAME
1214 if (lsi->lsi_flags & LSI_BDI_INITIALIZED) {
1215 bdi_destroy(&lsi->lsi_bdi);
1216 lsi->lsi_flags &= ~LSI_BDI_INITIALIZED;
1221 lsi->lsi_llsbi = NULL;
1223 lustre_common_put_super(sb);
1225 cl_env_cache_purge(~0);
1227 module_put(THIS_MODULE);
1230 } /* client_put_super */
1232 struct inode *ll_inode_from_resource_lock(struct ldlm_lock *lock)
1234 struct inode *inode = NULL;
1236 /* NOTE: we depend on atomic igrab() -bzzz */
1237 lock_res_and_lock(lock);
1238 if (lock->l_resource->lr_lvb_inode) {
1239 struct ll_inode_info * lli;
1240 lli = ll_i2info(lock->l_resource->lr_lvb_inode);
1241 if (lli->lli_inode_magic == LLI_INODE_MAGIC) {
1242 inode = igrab(lock->l_resource->lr_lvb_inode);
1244 inode = lock->l_resource->lr_lvb_inode;
1245 LDLM_DEBUG_LIMIT(inode->i_state & I_FREEING ? D_INFO :
1246 D_WARNING, lock, "lr_lvb_inode %p is "
1247 "bogus: magic %08x",
1248 lock->l_resource->lr_lvb_inode,
1249 lli->lli_inode_magic);
1253 unlock_res_and_lock(lock);
1257 void ll_dir_clear_lsm_md(struct inode *inode)
1259 struct ll_inode_info *lli = ll_i2info(inode);
1261 LASSERT(S_ISDIR(inode->i_mode));
1263 if (lli->lli_lsm_md != NULL) {
1264 lmv_free_memmd(lli->lli_lsm_md);
1265 lli->lli_lsm_md = NULL;
1269 static struct inode *ll_iget_anon_dir(struct super_block *sb,
1270 const struct lu_fid *fid,
1271 struct lustre_md *md)
1273 struct ll_sb_info *sbi = ll_s2sbi(sb);
1274 struct mdt_body *body = md->body;
1275 struct inode *inode;
1279 ino = cl_fid_build_ino(fid, sbi->ll_flags & LL_SBI_32BIT_API);
1280 inode = iget_locked(sb, ino);
1281 if (inode == NULL) {
1282 CERROR("%s: failed get simple inode "DFID": rc = -ENOENT\n",
1283 ll_get_fsname(sb, NULL, 0), PFID(fid));
1284 RETURN(ERR_PTR(-ENOENT));
1287 if (inode->i_state & I_NEW) {
1288 struct ll_inode_info *lli = ll_i2info(inode);
1289 struct lmv_stripe_md *lsm = md->lmv;
1291 inode->i_mode = (inode->i_mode & ~S_IFMT) |
1292 (body->mbo_mode & S_IFMT);
1293 LASSERTF(S_ISDIR(inode->i_mode), "Not slave inode "DFID"\n",
1296 inode->i_mtime.tv_sec = 0;
1297 inode->i_atime.tv_sec = 0;
1298 inode->i_ctime.tv_sec = 0;
1301 #ifdef HAVE_BACKING_DEV_INFO
1302 /* initializing backing dev info. */
1303 inode->i_mapping->backing_dev_info =
1304 &s2lsi(inode->i_sb)->lsi_bdi;
1306 inode->i_op = &ll_dir_inode_operations;
1307 inode->i_fop = &ll_dir_operations;
1308 lli->lli_fid = *fid;
1311 LASSERT(lsm != NULL);
1312 /* master object FID */
1313 lli->lli_pfid = body->mbo_fid1;
1314 CDEBUG(D_INODE, "lli %p slave "DFID" master "DFID"\n",
1315 lli, PFID(fid), PFID(&lli->lli_pfid));
1316 unlock_new_inode(inode);
1322 static int ll_init_lsm_md(struct inode *inode, struct lustre_md *md)
1325 struct lmv_stripe_md *lsm = md->lmv;
1326 struct ll_inode_info *lli = ll_i2info(inode);
1329 LASSERT(lsm != NULL);
1331 CDEBUG(D_INODE, "%s: "DFID" set dir layout:\n",
1332 ll_get_fsname(inode->i_sb, NULL, 0),
1333 PFID(&lli->lli_fid));
1334 lsm_md_dump(D_INODE, lsm);
1336 /* XXX sigh, this lsm_root initialization should be in
1337 * LMV layer, but it needs ll_iget right now, so we
1338 * put this here right now. */
1339 for (i = 0; i < lsm->lsm_md_stripe_count; i++) {
1340 fid = &lsm->lsm_md_oinfo[i].lmo_fid;
1341 LASSERT(lsm->lsm_md_oinfo[i].lmo_root == NULL);
1342 /* Unfortunately ll_iget will call ll_update_inode,
1343 * where the initialization of slave inode is slightly
1344 * different, so it reset lsm_md to NULL to avoid
1345 * initializing lsm for slave inode. */
1346 lsm->lsm_md_oinfo[i].lmo_root =
1347 ll_iget_anon_dir(inode->i_sb, fid, md);
1348 if (IS_ERR(lsm->lsm_md_oinfo[i].lmo_root)) {
1349 int rc = PTR_ERR(lsm->lsm_md_oinfo[i].lmo_root);
1351 lsm->lsm_md_oinfo[i].lmo_root = NULL;
1353 iput(lsm->lsm_md_oinfo[i].lmo_root);
1354 lsm->lsm_md_oinfo[i].lmo_root = NULL;
1360 lli->lli_lsm_md = lsm;
1365 static int ll_update_lsm_md(struct inode *inode, struct lustre_md *md)
1367 struct ll_inode_info *lli = ll_i2info(inode);
1368 struct lmv_stripe_md *lsm = md->lmv;
1373 LASSERT(S_ISDIR(inode->i_mode));
1374 CDEBUG(D_INODE, "update lsm %p of "DFID"\n", lli->lli_lsm_md,
1375 PFID(ll_inode2fid(inode)));
1378 * no striped information from request, lustre_md from req does not
1379 * include stripeEA, see ll_md_setattr()
1385 * normally dir layout doesn't change, only take read lock to check
1386 * that to avoid blocking other MD operations.
1388 if (lli->lli_lsm_md)
1389 down_read(&lli->lli_lsm_sem);
1391 down_write(&lli->lli_lsm_sem);
1394 * if dir layout mismatch, check whether version is increased, which
1395 * means layout is changed, this happens in dir migration and lfsck.
1397 if (lli->lli_lsm_md && !lsm_md_eq(lli->lli_lsm_md, lsm)) {
1398 if (lsm->lsm_md_layout_version <=
1399 lli->lli_lsm_md->lsm_md_layout_version) {
1400 CERROR("%s: "DFID" dir layout mismatch:\n",
1401 ll_get_fsname(inode->i_sb, NULL, 0),
1402 PFID(&lli->lli_fid));
1403 lsm_md_dump(D_ERROR, lli->lli_lsm_md);
1404 lsm_md_dump(D_ERROR, lsm);
1405 GOTO(unlock, rc = -EINVAL);
1408 /* layout changed, switch to write lock */
1409 up_read(&lli->lli_lsm_sem);
1410 down_write(&lli->lli_lsm_sem);
1411 ll_dir_clear_lsm_md(inode);
1414 /* set directory layout */
1415 if (!lli->lli_lsm_md) {
1416 struct cl_attr *attr;
1418 rc = ll_init_lsm_md(inode, md);
1419 up_write(&lli->lli_lsm_sem);
1423 /* set md->lmv to NULL, so the following free lustre_md
1424 * will not free this lsm */
1428 * md_merge_attr() may take long, since lsm is already set,
1429 * switch to read lock.
1431 down_read(&lli->lli_lsm_sem);
1433 OBD_ALLOC_PTR(attr);
1435 GOTO(unlock, rc = -ENOMEM);
1437 /* validate the lsm */
1438 rc = md_merge_attr(ll_i2mdexp(inode), lsm, attr,
1439 ll_md_blocking_ast);
1445 if (md->body->mbo_valid & OBD_MD_FLNLINK)
1446 md->body->mbo_nlink = attr->cat_nlink;
1447 if (md->body->mbo_valid & OBD_MD_FLSIZE)
1448 md->body->mbo_size = attr->cat_size;
1449 if (md->body->mbo_valid & OBD_MD_FLATIME)
1450 md->body->mbo_atime = attr->cat_atime;
1451 if (md->body->mbo_valid & OBD_MD_FLCTIME)
1452 md->body->mbo_ctime = attr->cat_ctime;
1453 if (md->body->mbo_valid & OBD_MD_FLMTIME)
1454 md->body->mbo_mtime = attr->cat_mtime;
1459 up_read(&lli->lli_lsm_sem);
1464 void ll_clear_inode(struct inode *inode)
1466 struct ll_inode_info *lli = ll_i2info(inode);
1467 struct ll_sb_info *sbi = ll_i2sbi(inode);
1470 CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p)\n",
1471 PFID(ll_inode2fid(inode)), inode);
1473 if (S_ISDIR(inode->i_mode)) {
1474 /* these should have been cleared in ll_file_release */
1475 LASSERT(lli->lli_opendir_key == NULL);
1476 LASSERT(lli->lli_sai == NULL);
1477 LASSERT(lli->lli_opendir_pid == 0);
1480 md_null_inode(sbi->ll_md_exp, ll_inode2fid(inode));
1482 LASSERT(!lli->lli_open_fd_write_count);
1483 LASSERT(!lli->lli_open_fd_read_count);
1484 LASSERT(!lli->lli_open_fd_exec_count);
1486 if (lli->lli_mds_write_och)
1487 ll_md_real_close(inode, FMODE_WRITE);
1488 if (lli->lli_mds_exec_och)
1489 ll_md_real_close(inode, FMODE_EXEC);
1490 if (lli->lli_mds_read_och)
1491 ll_md_real_close(inode, FMODE_READ);
1493 if (S_ISLNK(inode->i_mode) && lli->lli_symlink_name) {
1494 OBD_FREE(lli->lli_symlink_name,
1495 strlen(lli->lli_symlink_name) + 1);
1496 lli->lli_symlink_name = NULL;
1499 ll_xattr_cache_destroy(inode);
1501 #ifdef CONFIG_FS_POSIX_ACL
1502 forget_all_cached_acls(inode);
1503 if (lli->lli_posix_acl) {
1504 posix_acl_release(lli->lli_posix_acl);
1505 lli->lli_posix_acl = NULL;
1508 lli->lli_inode_magic = LLI_INODE_DEAD;
1510 if (S_ISDIR(inode->i_mode))
1511 ll_dir_clear_lsm_md(inode);
1512 else if (S_ISREG(inode->i_mode) && !is_bad_inode(inode))
1513 LASSERT(list_empty(&lli->lli_agl_list));
1516 * XXX This has to be done before lsm is freed below, because
1517 * cl_object still uses inode lsm.
1519 cl_inode_fini(inode);
1524 static int ll_md_setattr(struct dentry *dentry, struct md_op_data *op_data)
1526 struct lustre_md md;
1527 struct inode *inode = dentry->d_inode;
1528 struct ll_sb_info *sbi = ll_i2sbi(inode);
1529 struct ptlrpc_request *request = NULL;
1533 op_data = ll_prep_md_op_data(op_data, inode, NULL, NULL, 0, 0,
1534 LUSTRE_OPC_ANY, NULL);
1535 if (IS_ERR(op_data))
1536 RETURN(PTR_ERR(op_data));
1538 rc = md_setattr(sbi->ll_md_exp, op_data, NULL, 0, &request);
1540 ptlrpc_req_finished(request);
1541 if (rc == -ENOENT) {
1543 /* Unlinked special device node? Or just a race?
1544 * Pretend we done everything. */
1545 if (!S_ISREG(inode->i_mode) &&
1546 !S_ISDIR(inode->i_mode)) {
1547 ia_valid = op_data->op_attr.ia_valid;
1548 op_data->op_attr.ia_valid &= ~TIMES_SET_FLAGS;
1549 rc = simple_setattr(dentry, &op_data->op_attr);
1550 op_data->op_attr.ia_valid = ia_valid;
1552 } else if (rc != -EPERM && rc != -EACCES && rc != -ETXTBSY) {
1553 CERROR("md_setattr fails: rc = %d\n", rc);
1558 rc = md_get_lustre_md(sbi->ll_md_exp, request, sbi->ll_dt_exp,
1559 sbi->ll_md_exp, &md);
1561 ptlrpc_req_finished(request);
1565 ia_valid = op_data->op_attr.ia_valid;
1566 /* inode size will be in ll_setattr_ost, can't do it now since dirty
1567 * cache is not cleared yet. */
1568 op_data->op_attr.ia_valid &= ~(TIMES_SET_FLAGS | ATTR_SIZE);
1569 if (S_ISREG(inode->i_mode))
1571 rc = simple_setattr(dentry, &op_data->op_attr);
1572 if (S_ISREG(inode->i_mode))
1573 inode_unlock(inode);
1574 op_data->op_attr.ia_valid = ia_valid;
1576 rc = ll_update_inode(inode, &md);
1577 ptlrpc_req_finished(request);
1582 /* If this inode has objects allocated to it (lsm != NULL), then the OST
1583 * object(s) determine the file size and mtime. Otherwise, the MDS will
1584 * keep these values until such a time that objects are allocated for it.
1585 * We do the MDS operations first, as it is checking permissions for us.
1586 * We don't to the MDS RPC if there is nothing that we want to store there,
1587 * otherwise there is no harm in updating mtime/atime on the MDS if we are
1588 * going to do an RPC anyways.
1590 * If we are doing a truncate, we will send the mtime and ctime updates
1591 * to the OST with the punch RPC, otherwise we do an explicit setattr RPC.
1592 * I don't believe it is possible to get e.g. ATTR_MTIME_SET and ATTR_SIZE
1595 * In case of HSMimport, we only set attr on MDS.
1597 int ll_setattr_raw(struct dentry *dentry, struct iattr *attr,
1598 enum op_xvalid xvalid, bool hsm_import)
1600 struct inode *inode = dentry->d_inode;
1601 struct ll_inode_info *lli = ll_i2info(inode);
1602 struct md_op_data *op_data = NULL;
1606 CDEBUG(D_VFSTRACE, "%s: setattr inode "DFID"(%p) from %llu to %llu, "
1607 "valid %x, hsm_import %d\n",
1608 ll_get_fsname(inode->i_sb, NULL, 0), PFID(&lli->lli_fid),
1609 inode, i_size_read(inode), attr->ia_size, attr->ia_valid,
1612 if (attr->ia_valid & ATTR_SIZE) {
1613 /* Check new size against VFS/VM file size limit and rlimit */
1614 rc = inode_newsize_ok(inode, attr->ia_size);
1618 /* The maximum Lustre file size is variable, based on the
1619 * OST maximum object size and number of stripes. This
1620 * needs another check in addition to the VFS check above. */
1621 if (attr->ia_size > ll_file_maxbytes(inode)) {
1622 CDEBUG(D_INODE,"file "DFID" too large %llu > %llu\n",
1623 PFID(&lli->lli_fid), attr->ia_size,
1624 ll_file_maxbytes(inode));
1628 attr->ia_valid |= ATTR_MTIME | ATTR_CTIME;
1631 /* POSIX: check before ATTR_*TIME_SET set (from inode_change_ok) */
1632 if (attr->ia_valid & TIMES_SET_FLAGS) {
1633 if ((!uid_eq(current_fsuid(), inode->i_uid)) &&
1634 !cfs_capable(CFS_CAP_FOWNER))
1638 /* We mark all of the fields "set" so MDS/OST does not re-set them */
1639 if (!(xvalid & OP_XVALID_CTIME_SET) &&
1640 (attr->ia_valid & ATTR_CTIME)) {
1641 attr->ia_ctime = current_time(inode);
1642 xvalid |= OP_XVALID_CTIME_SET;
1644 if (!(attr->ia_valid & ATTR_ATIME_SET) &&
1645 (attr->ia_valid & ATTR_ATIME)) {
1646 attr->ia_atime = current_time(inode);
1647 attr->ia_valid |= ATTR_ATIME_SET;
1649 if (!(attr->ia_valid & ATTR_MTIME_SET) &&
1650 (attr->ia_valid & ATTR_MTIME)) {
1651 attr->ia_mtime = current_time(inode);
1652 attr->ia_valid |= ATTR_MTIME_SET;
1655 if (attr->ia_valid & (ATTR_MTIME | ATTR_CTIME))
1656 CDEBUG(D_INODE, "setting mtime %lld, ctime %lld, now = %lld\n",
1657 (s64)attr->ia_mtime.tv_sec, (s64)attr->ia_ctime.tv_sec,
1658 ktime_get_real_seconds());
1660 if (S_ISREG(inode->i_mode)) {
1661 if (attr->ia_valid & ATTR_SIZE)
1662 inode_dio_write_done(inode);
1663 inode_unlock(inode);
1666 /* We always do an MDS RPC, even if we're only changing the size;
1667 * only the MDS knows whether truncate() should fail with -ETXTBUSY */
1669 OBD_ALLOC_PTR(op_data);
1670 if (op_data == NULL)
1671 GOTO(out, rc = -ENOMEM);
1673 if (!hsm_import && attr->ia_valid & ATTR_SIZE) {
1674 /* If we are changing file size, file content is
1675 * modified, flag it.
1677 xvalid |= OP_XVALID_OWNEROVERRIDE;
1678 op_data->op_bias |= MDS_DATA_MODIFIED;
1679 ll_file_clear_flag(lli, LLIF_DATA_MODIFIED);
1682 if (attr->ia_valid & ATTR_FILE) {
1683 struct ll_file_data *fd = LUSTRE_FPRIVATE(attr->ia_file);
1685 if (fd->fd_lease_och)
1686 op_data->op_bias |= MDS_TRUNC_KEEP_LEASE;
1689 op_data->op_attr = *attr;
1690 op_data->op_xvalid = xvalid;
1692 rc = ll_md_setattr(dentry, op_data);
1696 if (!S_ISREG(inode->i_mode) || hsm_import)
1699 if (attr->ia_valid & (ATTR_SIZE | ATTR_ATIME | ATTR_ATIME_SET |
1700 ATTR_MTIME | ATTR_MTIME_SET | ATTR_CTIME) ||
1701 xvalid & OP_XVALID_CTIME_SET) {
1702 /* For truncate and utimes sending attributes to OSTs, setting
1703 * mtime/atime to the past will be performed under PW [0:EOF]
1704 * extent lock (new_size:EOF for truncate). It may seem
1705 * excessive to send mtime/atime updates to OSTs when not
1706 * setting times to past, but it is necessary due to possible
1707 * time de-synchronization between MDT inode and OST objects
1709 rc = cl_setattr_ost(lli->lli_clob, attr, xvalid, 0);
1712 /* If the file was restored, it needs to set dirty flag.
1714 * We've already sent MDS_DATA_MODIFIED flag in
1715 * ll_md_setattr() for truncate. However, the MDT refuses to
1716 * set the HS_DIRTY flag on released files, so we have to set
1717 * it again if the file has been restored. Please check how
1718 * LLIF_DATA_MODIFIED is set in vvp_io_setattr_fini().
1720 * Please notice that if the file is not released, the previous
1721 * MDS_DATA_MODIFIED has taken effect and usually
1722 * LLIF_DATA_MODIFIED is not set(see vvp_io_setattr_fini()).
1723 * This way we can save an RPC for common open + trunc
1725 if (ll_file_test_and_clear_flag(lli, LLIF_DATA_MODIFIED)) {
1726 struct hsm_state_set hss = {
1727 .hss_valid = HSS_SETMASK,
1728 .hss_setmask = HS_DIRTY,
1732 rc2 = ll_hsm_state_set(inode, &hss);
1733 /* truncate and write can happen at the same time, so that
1734 * the file can be set modified even though the file is not
1735 * restored from released state, and ll_hsm_state_set() is
1736 * not applicable for the file, and rc2 < 0 is normal in this
1739 CDEBUG(D_INFO, DFID "HSM set dirty failed: rc2 = %d\n",
1740 PFID(ll_inode2fid(inode)), rc2);
1745 if (op_data != NULL)
1746 ll_finish_md_op_data(op_data);
1748 if (S_ISREG(inode->i_mode)) {
1750 if ((attr->ia_valid & ATTR_SIZE) && !hsm_import)
1751 inode_dio_wait(inode);
1752 /* Once we've got the i_mutex, it's safe to set the S_NOSEC
1753 * flag. ll_update_inode (called from ll_md_setattr), clears
1754 * inode flags, so there is a gap where S_NOSEC is not set.
1755 * This can cause a writer to take the i_mutex unnecessarily,
1756 * but this is safe to do and should be rare. */
1757 inode_has_no_xattr(inode);
1760 ll_stats_ops_tally(ll_i2sbi(inode), (attr->ia_valid & ATTR_SIZE) ?
1761 LPROC_LL_TRUNC : LPROC_LL_SETATTR, 1);
1766 int ll_setattr(struct dentry *de, struct iattr *attr)
1768 int mode = de->d_inode->i_mode;
1769 enum op_xvalid xvalid = 0;
1771 if ((attr->ia_valid & (ATTR_CTIME|ATTR_SIZE|ATTR_MODE)) ==
1772 (ATTR_CTIME|ATTR_SIZE|ATTR_MODE))
1773 xvalid |= OP_XVALID_OWNEROVERRIDE;
1775 if (((attr->ia_valid & (ATTR_MODE|ATTR_FORCE|ATTR_SIZE)) ==
1776 (ATTR_SIZE|ATTR_MODE)) &&
1777 (((mode & S_ISUID) && !(attr->ia_mode & S_ISUID)) ||
1778 (((mode & (S_ISGID|S_IXGRP)) == (S_ISGID|S_IXGRP)) &&
1779 !(attr->ia_mode & S_ISGID))))
1780 attr->ia_valid |= ATTR_FORCE;
1782 if ((attr->ia_valid & ATTR_MODE) &&
1784 !(attr->ia_mode & S_ISUID) &&
1785 !(attr->ia_valid & ATTR_KILL_SUID))
1786 attr->ia_valid |= ATTR_KILL_SUID;
1788 if ((attr->ia_valid & ATTR_MODE) &&
1789 ((mode & (S_ISGID|S_IXGRP)) == (S_ISGID|S_IXGRP)) &&
1790 !(attr->ia_mode & S_ISGID) &&
1791 !(attr->ia_valid & ATTR_KILL_SGID))
1792 attr->ia_valid |= ATTR_KILL_SGID;
1794 return ll_setattr_raw(de, attr, xvalid, false);
1797 int ll_statfs_internal(struct ll_sb_info *sbi, struct obd_statfs *osfs,
1800 struct obd_statfs obd_osfs = { 0 };
1805 max_age = ktime_get_seconds() - OBD_STATFS_CACHE_SECONDS;
1807 rc = obd_statfs(NULL, sbi->ll_md_exp, osfs, max_age, flags);
1811 osfs->os_type = LL_SUPER_MAGIC;
1813 CDEBUG(D_SUPER, "MDC blocks %llu/%llu objects %llu/%llu\n",
1814 osfs->os_bavail, osfs->os_blocks, osfs->os_ffree, osfs->os_files);
1816 if (osfs->os_state & OS_STATE_SUM)
1819 if (sbi->ll_flags & LL_SBI_LAZYSTATFS)
1820 flags |= OBD_STATFS_NODELAY;
1822 rc = obd_statfs(NULL, sbi->ll_dt_exp, &obd_osfs, max_age, flags);
1823 if (rc) /* Possibly a filesystem with no OSTs. Report MDT totals. */
1826 CDEBUG(D_SUPER, "OSC blocks %llu/%llu objects %llu/%llu\n",
1827 obd_osfs.os_bavail, obd_osfs.os_blocks, obd_osfs.os_ffree,
1830 osfs->os_bsize = obd_osfs.os_bsize;
1831 osfs->os_blocks = obd_osfs.os_blocks;
1832 osfs->os_bfree = obd_osfs.os_bfree;
1833 osfs->os_bavail = obd_osfs.os_bavail;
1835 /* If we have _some_ OSTs, but don't have as many free objects on the
1836 * OSTs as inodes on the MDTs, reduce the reported number of inodes
1837 * to compensate, so that the "inodes in use" number is correct.
1838 * This should be kept in sync with lod_statfs() behaviour.
1840 if (obd_osfs.os_files && obd_osfs.os_ffree < osfs->os_ffree) {
1841 osfs->os_files = (osfs->os_files - osfs->os_ffree) +
1843 osfs->os_ffree = obd_osfs.os_ffree;
1849 int ll_statfs(struct dentry *de, struct kstatfs *sfs)
1851 struct super_block *sb = de->d_sb;
1852 struct obd_statfs osfs;
1853 __u64 fsid = huge_encode_dev(sb->s_dev);
1856 CDEBUG(D_VFSTRACE, "VFS Op: at %llu jiffies\n", get_jiffies_64());
1857 ll_stats_ops_tally(ll_s2sbi(sb), LPROC_LL_STAFS, 1);
1859 /* Some amount of caching on the client is allowed */
1860 rc = ll_statfs_internal(ll_s2sbi(sb), &osfs, OBD_STATFS_SUM);
1864 statfs_unpack(sfs, &osfs);
1866 /* We need to downshift for all 32-bit kernels, because we can't
1867 * tell if the kernel is being called via sys_statfs64() or not.
1868 * Stop before overflowing f_bsize - in which case it is better
1869 * to just risk EOVERFLOW if caller is using old sys_statfs(). */
1870 if (sizeof(long) < 8) {
1871 while (osfs.os_blocks > ~0UL && sfs->f_bsize < 0x40000000) {
1874 osfs.os_blocks >>= 1;
1875 osfs.os_bfree >>= 1;
1876 osfs.os_bavail >>= 1;
1880 sfs->f_blocks = osfs.os_blocks;
1881 sfs->f_bfree = osfs.os_bfree;
1882 sfs->f_bavail = osfs.os_bavail;
1883 sfs->f_fsid.val[0] = (__u32)fsid;
1884 sfs->f_fsid.val[1] = (__u32)(fsid >> 32);
1888 void ll_inode_size_lock(struct inode *inode)
1890 struct ll_inode_info *lli;
1892 LASSERT(!S_ISDIR(inode->i_mode));
1894 lli = ll_i2info(inode);
1895 mutex_lock(&lli->lli_size_mutex);
1898 void ll_inode_size_unlock(struct inode *inode)
1900 struct ll_inode_info *lli;
1902 lli = ll_i2info(inode);
1903 mutex_unlock(&lli->lli_size_mutex);
1906 void ll_update_inode_flags(struct inode *inode, int ext_flags)
1908 inode->i_flags = ll_ext_to_inode_flags(ext_flags);
1909 if (ext_flags & LUSTRE_PROJINHERIT_FL)
1910 ll_file_set_flag(ll_i2info(inode), LLIF_PROJECT_INHERIT);
1912 ll_file_clear_flag(ll_i2info(inode), LLIF_PROJECT_INHERIT);
1915 int ll_update_inode(struct inode *inode, struct lustre_md *md)
1917 struct ll_inode_info *lli = ll_i2info(inode);
1918 struct mdt_body *body = md->body;
1919 struct ll_sb_info *sbi = ll_i2sbi(inode);
1922 if (body->mbo_valid & OBD_MD_FLEASIZE) {
1923 rc = cl_file_inode_init(inode, md);
1928 if (S_ISDIR(inode->i_mode)) {
1929 rc = ll_update_lsm_md(inode, md);
1934 #ifdef CONFIG_FS_POSIX_ACL
1935 if (body->mbo_valid & OBD_MD_FLACL) {
1936 spin_lock(&lli->lli_lock);
1937 if (lli->lli_posix_acl)
1938 posix_acl_release(lli->lli_posix_acl);
1939 lli->lli_posix_acl = md->posix_acl;
1940 spin_unlock(&lli->lli_lock);
1943 inode->i_ino = cl_fid_build_ino(&body->mbo_fid1,
1944 sbi->ll_flags & LL_SBI_32BIT_API);
1945 inode->i_generation = cl_fid_build_gen(&body->mbo_fid1);
1947 if (body->mbo_valid & OBD_MD_FLATIME) {
1948 if (body->mbo_atime > inode->i_atime.tv_sec)
1949 inode->i_atime.tv_sec = body->mbo_atime;
1950 lli->lli_atime = body->mbo_atime;
1953 if (body->mbo_valid & OBD_MD_FLMTIME) {
1954 if (body->mbo_mtime > inode->i_mtime.tv_sec) {
1956 "setting ino %lu mtime from %lld to %llu\n",
1957 inode->i_ino, (s64)inode->i_mtime.tv_sec,
1959 inode->i_mtime.tv_sec = body->mbo_mtime;
1961 lli->lli_mtime = body->mbo_mtime;
1964 if (body->mbo_valid & OBD_MD_FLCTIME) {
1965 if (body->mbo_ctime > inode->i_ctime.tv_sec)
1966 inode->i_ctime.tv_sec = body->mbo_ctime;
1967 lli->lli_ctime = body->mbo_ctime;
1970 /* Clear i_flags to remove S_NOSEC before permissions are updated */
1971 if (body->mbo_valid & OBD_MD_FLFLAGS)
1972 ll_update_inode_flags(inode, body->mbo_flags);
1973 if (body->mbo_valid & OBD_MD_FLMODE)
1974 inode->i_mode = (inode->i_mode & S_IFMT) |
1975 (body->mbo_mode & ~S_IFMT);
1977 if (body->mbo_valid & OBD_MD_FLTYPE)
1978 inode->i_mode = (inode->i_mode & ~S_IFMT) |
1979 (body->mbo_mode & S_IFMT);
1981 LASSERT(inode->i_mode != 0);
1982 if (S_ISREG(inode->i_mode))
1983 inode->i_blkbits = min(PTLRPC_MAX_BRW_BITS + 1,
1984 LL_MAX_BLKSIZE_BITS);
1986 inode->i_blkbits = inode->i_sb->s_blocksize_bits;
1988 if (body->mbo_valid & OBD_MD_FLUID)
1989 inode->i_uid = make_kuid(&init_user_ns, body->mbo_uid);
1990 if (body->mbo_valid & OBD_MD_FLGID)
1991 inode->i_gid = make_kgid(&init_user_ns, body->mbo_gid);
1992 if (body->mbo_valid & OBD_MD_FLPROJID)
1993 lli->lli_projid = body->mbo_projid;
1994 if (body->mbo_valid & OBD_MD_FLNLINK)
1995 set_nlink(inode, body->mbo_nlink);
1996 if (body->mbo_valid & OBD_MD_FLRDEV)
1997 inode->i_rdev = old_decode_dev(body->mbo_rdev);
1999 if (body->mbo_valid & OBD_MD_FLID) {
2000 /* FID shouldn't be changed! */
2001 if (fid_is_sane(&lli->lli_fid)) {
2002 LASSERTF(lu_fid_eq(&lli->lli_fid, &body->mbo_fid1),
2003 "Trying to change FID "DFID
2004 " to the "DFID", inode "DFID"(%p)\n",
2005 PFID(&lli->lli_fid), PFID(&body->mbo_fid1),
2006 PFID(ll_inode2fid(inode)), inode);
2008 lli->lli_fid = body->mbo_fid1;
2012 LASSERT(fid_seq(&lli->lli_fid) != 0);
2014 if (body->mbo_valid & OBD_MD_FLSIZE) {
2015 i_size_write(inode, body->mbo_size);
2017 CDEBUG(D_VFSTRACE, "inode="DFID", updating i_size %llu\n",
2018 PFID(ll_inode2fid(inode)),
2019 (unsigned long long)body->mbo_size);
2021 if (body->mbo_valid & OBD_MD_FLBLOCKS)
2022 inode->i_blocks = body->mbo_blocks;
2025 if (body->mbo_valid & OBD_MD_TSTATE) {
2026 /* Set LLIF_FILE_RESTORING if restore ongoing and
2027 * clear it when done to ensure to start again
2028 * glimpsing updated attrs
2030 if (body->mbo_t_state & MS_RESTORE)
2031 ll_file_set_flag(lli, LLIF_FILE_RESTORING);
2033 ll_file_clear_flag(lli, LLIF_FILE_RESTORING);
2039 int ll_read_inode2(struct inode *inode, void *opaque)
2041 struct lustre_md *md = opaque;
2042 struct ll_inode_info *lli = ll_i2info(inode);
2046 CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p)\n",
2047 PFID(&lli->lli_fid), inode);
2049 /* Core attributes from the MDS first. This is a new inode, and
2050 * the VFS doesn't zero times in the core inode so we have to do
2051 * it ourselves. They will be overwritten by either MDS or OST
2052 * attributes - we just need to make sure they aren't newer.
2054 inode->i_mtime.tv_sec = 0;
2055 inode->i_atime.tv_sec = 0;
2056 inode->i_ctime.tv_sec = 0;
2058 rc = ll_update_inode(inode, md);
2062 /* OIDEBUG(inode); */
2064 #ifdef HAVE_BACKING_DEV_INFO
2065 /* initializing backing dev info. */
2066 inode->i_mapping->backing_dev_info = &s2lsi(inode->i_sb)->lsi_bdi;
2068 if (S_ISREG(inode->i_mode)) {
2069 struct ll_sb_info *sbi = ll_i2sbi(inode);
2070 inode->i_op = &ll_file_inode_operations;
2071 inode->i_fop = sbi->ll_fop;
2072 inode->i_mapping->a_ops = (struct address_space_operations *)&ll_aops;
2074 } else if (S_ISDIR(inode->i_mode)) {
2075 inode->i_op = &ll_dir_inode_operations;
2076 inode->i_fop = &ll_dir_operations;
2078 } else if (S_ISLNK(inode->i_mode)) {
2079 inode->i_op = &ll_fast_symlink_inode_operations;
2082 inode->i_op = &ll_special_inode_operations;
2084 init_special_inode(inode, inode->i_mode,
2093 void ll_delete_inode(struct inode *inode)
2095 struct ll_inode_info *lli = ll_i2info(inode);
2096 struct address_space *mapping = &inode->i_data;
2097 unsigned long nrpages;
2100 if (S_ISREG(inode->i_mode) && lli->lli_clob != NULL)
2101 /* It is last chance to write out dirty pages,
2102 * otherwise we may lose data while umount */
2103 cl_sync_file_range(inode, 0, OBD_OBJECT_EOF, CL_FSYNC_LOCAL, 1);
2105 truncate_inode_pages_final(mapping);
2107 /* Workaround for LU-118: Note nrpages may not be totally updated when
2108 * truncate_inode_pages() returns, as there can be a page in the process
2109 * of deletion (inside __delete_from_page_cache()) in the specified
2110 * range. Thus mapping->nrpages can be non-zero when this function
2111 * returns even after truncation of the whole mapping. Only do this if
2112 * npages isn't already zero.
2114 nrpages = mapping->nrpages;
2116 spin_lock_irq(&mapping->tree_lock);
2117 nrpages = mapping->nrpages;
2118 spin_unlock_irq(&mapping->tree_lock);
2119 } /* Workaround end */
2121 LASSERTF(nrpages == 0, "%s: inode="DFID"(%p) nrpages=%lu, "
2122 "see https://jira.whamcloud.com/browse/LU-118\n",
2123 ll_get_fsname(inode->i_sb, NULL, 0),
2124 PFID(ll_inode2fid(inode)), inode, nrpages);
2126 #ifdef HAVE_SBOPS_EVICT_INODE
2127 ll_clear_inode(inode);
2134 int ll_iocontrol(struct inode *inode, struct file *file,
2135 unsigned int cmd, unsigned long arg)
2137 struct ll_sb_info *sbi = ll_i2sbi(inode);
2138 struct ptlrpc_request *req = NULL;
2143 case FS_IOC_GETFLAGS: {
2144 struct mdt_body *body;
2145 struct md_op_data *op_data;
2147 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL,
2148 0, 0, LUSTRE_OPC_ANY,
2150 if (IS_ERR(op_data))
2151 RETURN(PTR_ERR(op_data));
2153 op_data->op_valid = OBD_MD_FLFLAGS;
2154 rc = md_getattr(sbi->ll_md_exp, op_data, &req);
2155 ll_finish_md_op_data(op_data);
2157 CERROR("%s: failure inode "DFID": rc = %d\n",
2158 sbi->ll_md_exp->exp_obd->obd_name,
2159 PFID(ll_inode2fid(inode)), rc);
2163 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
2165 flags = body->mbo_flags;
2167 ptlrpc_req_finished(req);
2169 RETURN(put_user(flags, (int __user *)arg));
2171 case FS_IOC_SETFLAGS: {
2173 struct md_op_data *op_data;
2174 struct cl_object *obj;
2175 struct fsxattr fa = { 0 };
2177 if (get_user(flags, (int __user *)arg))
2180 fa.fsx_projid = ll_i2info(inode)->lli_projid;
2181 if (flags & LUSTRE_PROJINHERIT_FL)
2182 fa.fsx_xflags = FS_XFLAG_PROJINHERIT;
2184 rc = ll_ioctl_check_project(inode, &fa);
2188 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
2189 LUSTRE_OPC_ANY, NULL);
2190 if (IS_ERR(op_data))
2191 RETURN(PTR_ERR(op_data));
2193 op_data->op_attr_flags = flags;
2194 op_data->op_xvalid |= OP_XVALID_FLAGS;
2195 rc = md_setattr(sbi->ll_md_exp, op_data, NULL, 0, &req);
2196 ll_finish_md_op_data(op_data);
2197 ptlrpc_req_finished(req);
2201 ll_update_inode_flags(inode, flags);
2203 obj = ll_i2info(inode)->lli_clob;
2207 OBD_ALLOC_PTR(attr);
2211 rc = cl_setattr_ost(obj, attr, OP_XVALID_FLAGS, flags);
2223 int ll_flush_ctx(struct inode *inode)
2225 struct ll_sb_info *sbi = ll_i2sbi(inode);
2227 CDEBUG(D_SEC, "flush context for user %d\n",
2228 from_kuid(&init_user_ns, current_uid()));
2230 obd_set_info_async(NULL, sbi->ll_md_exp,
2231 sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
2233 obd_set_info_async(NULL, sbi->ll_dt_exp,
2234 sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
2239 /* umount -f client means force down, don't save state */
2240 void ll_umount_begin(struct super_block *sb)
2242 struct ll_sb_info *sbi = ll_s2sbi(sb);
2243 struct obd_device *obd;
2244 struct obd_ioctl_data *ioc_data;
2245 struct l_wait_info lwi;
2246 wait_queue_head_t waitq;
2249 CDEBUG(D_VFSTRACE, "VFS Op: superblock %p count %d active %d\n", sb,
2250 sb->s_count, atomic_read(&sb->s_active));
2252 obd = class_exp2obd(sbi->ll_md_exp);
2254 CERROR("Invalid MDC connection handle %#llx\n",
2255 sbi->ll_md_exp->exp_handle.h_cookie);
2261 obd = class_exp2obd(sbi->ll_dt_exp);
2263 CERROR("Invalid LOV connection handle %#llx\n",
2264 sbi->ll_dt_exp->exp_handle.h_cookie);
2270 OBD_ALLOC_PTR(ioc_data);
2272 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_md_exp,
2273 sizeof *ioc_data, ioc_data, NULL);
2275 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_dt_exp,
2276 sizeof *ioc_data, ioc_data, NULL);
2278 OBD_FREE_PTR(ioc_data);
2281 /* Really, we'd like to wait until there are no requests outstanding,
2282 * and then continue. For now, we just periodically checking for vfs
2283 * to decrement mnt_cnt and hope to finish it within 10sec.
2285 init_waitqueue_head(&waitq);
2286 lwi = LWI_TIMEOUT_INTERVAL(cfs_time_seconds(10),
2287 cfs_time_seconds(1), NULL, NULL);
2288 l_wait_event(waitq, may_umount(sbi->ll_mnt.mnt), &lwi);
2293 int ll_remount_fs(struct super_block *sb, int *flags, char *data)
2295 struct ll_sb_info *sbi = ll_s2sbi(sb);
2296 char *profilenm = get_profile_name(sb);
2300 if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY)) {
2301 read_only = *flags & MS_RDONLY;
2302 err = obd_set_info_async(NULL, sbi->ll_md_exp,
2303 sizeof(KEY_READ_ONLY),
2304 KEY_READ_ONLY, sizeof(read_only),
2307 LCONSOLE_WARN("Failed to remount %s %s (%d)\n",
2308 profilenm, read_only ?
2309 "read-only" : "read-write", err);
2314 sb->s_flags |= MS_RDONLY;
2316 sb->s_flags &= ~MS_RDONLY;
2318 if (sbi->ll_flags & LL_SBI_VERBOSE)
2319 LCONSOLE_WARN("Remounted %s %s\n", profilenm,
2320 read_only ? "read-only" : "read-write");
2326 * Cleanup the open handle that is cached on MDT-side.
2328 * For open case, the client side open handling thread may hit error
2329 * after the MDT grant the open. Under such case, the client should
2330 * send close RPC to the MDT as cleanup; otherwise, the open handle
2331 * on the MDT will be leaked there until the client umount or evicted.
2333 * In further, if someone unlinked the file, because the open handle
2334 * holds the reference on such file/object, then it will block the
2335 * subsequent threads that want to locate such object via FID.
2337 * \param[in] sb super block for this file-system
2338 * \param[in] open_req pointer to the original open request
2340 void ll_open_cleanup(struct super_block *sb, struct ptlrpc_request *open_req)
2342 struct mdt_body *body;
2343 struct md_op_data *op_data;
2344 struct ptlrpc_request *close_req = NULL;
2345 struct obd_export *exp = ll_s2sbi(sb)->ll_md_exp;
2348 body = req_capsule_server_get(&open_req->rq_pill, &RMF_MDT_BODY);
2349 OBD_ALLOC_PTR(op_data);
2350 if (op_data == NULL) {
2351 CWARN("%s: cannot allocate op_data to release open handle for "
2353 ll_get_fsname(sb, NULL, 0), PFID(&body->mbo_fid1));
2358 op_data->op_fid1 = body->mbo_fid1;
2359 op_data->op_open_handle = body->mbo_open_handle;
2360 op_data->op_mod_time = ktime_get_real_seconds();
2361 md_close(exp, op_data, NULL, &close_req);
2362 ptlrpc_req_finished(close_req);
2363 ll_finish_md_op_data(op_data);
2368 int ll_prep_inode(struct inode **inode, struct ptlrpc_request *req,
2369 struct super_block *sb, struct lookup_intent *it)
2371 struct ll_sb_info *sbi = NULL;
2372 struct lustre_md md = { NULL };
2376 LASSERT(*inode || sb);
2377 sbi = sb ? ll_s2sbi(sb) : ll_i2sbi(*inode);
2378 rc = md_get_lustre_md(sbi->ll_md_exp, req, sbi->ll_dt_exp,
2379 sbi->ll_md_exp, &md);
2384 rc = ll_update_inode(*inode, &md);
2388 LASSERT(sb != NULL);
2391 * At this point server returns to client's same fid as client
2392 * generated for creating. So using ->fid1 is okay here.
2394 if (!fid_is_sane(&md.body->mbo_fid1)) {
2395 CERROR("%s: Fid is insane "DFID"\n",
2396 ll_get_fsname(sb, NULL, 0),
2397 PFID(&md.body->mbo_fid1));
2398 GOTO(out, rc = -EINVAL);
2401 *inode = ll_iget(sb, cl_fid_build_ino(&md.body->mbo_fid1,
2402 sbi->ll_flags & LL_SBI_32BIT_API),
2404 if (IS_ERR(*inode)) {
2405 #ifdef CONFIG_FS_POSIX_ACL
2407 posix_acl_release(md.posix_acl);
2408 md.posix_acl = NULL;
2411 rc = IS_ERR(*inode) ? PTR_ERR(*inode) : -ENOMEM;
2413 CERROR("new_inode -fatal: rc %d\n", rc);
2418 /* Handling piggyback layout lock.
2419 * Layout lock can be piggybacked by getattr and open request.
2420 * The lsm can be applied to inode only if it comes with a layout lock
2421 * otherwise correct layout may be overwritten, for example:
2422 * 1. proc1: mdt returns a lsm but not granting layout
2423 * 2. layout was changed by another client
2424 * 3. proc2: refresh layout and layout lock granted
2425 * 4. proc1: to apply a stale layout */
2426 if (it != NULL && it->it_lock_mode != 0) {
2427 struct lustre_handle lockh;
2428 struct ldlm_lock *lock;
2430 lockh.cookie = it->it_lock_handle;
2431 lock = ldlm_handle2lock(&lockh);
2432 LASSERT(lock != NULL);
2433 if (ldlm_has_layout(lock)) {
2434 struct cl_object_conf conf;
2436 memset(&conf, 0, sizeof(conf));
2437 conf.coc_opc = OBJECT_CONF_SET;
2438 conf.coc_inode = *inode;
2439 conf.coc_lock = lock;
2440 conf.u.coc_layout = md.layout;
2441 (void)ll_layout_conf(*inode, &conf);
2443 LDLM_LOCK_PUT(lock);
2449 md_free_lustre_md(sbi->ll_md_exp, &md);
2452 if (rc != 0 && it != NULL && it->it_op & IT_OPEN)
2453 ll_open_cleanup(sb != NULL ? sb : (*inode)->i_sb, req);
2458 int ll_obd_statfs(struct inode *inode, void __user *arg)
2460 struct ll_sb_info *sbi = NULL;
2461 struct obd_export *exp;
2463 struct obd_ioctl_data *data = NULL;
2467 if (!inode || !(sbi = ll_i2sbi(inode)))
2468 GOTO(out_statfs, rc = -EINVAL);
2470 rc = obd_ioctl_getdata(&buf, &len, arg);
2472 GOTO(out_statfs, rc);
2475 if (!data->ioc_inlbuf1 || !data->ioc_inlbuf2 ||
2476 !data->ioc_pbuf1 || !data->ioc_pbuf2)
2477 GOTO(out_statfs, rc = -EINVAL);
2479 if (data->ioc_inllen1 != sizeof(__u32) ||
2480 data->ioc_inllen2 != sizeof(__u32) ||
2481 data->ioc_plen1 != sizeof(struct obd_statfs) ||
2482 data->ioc_plen2 != sizeof(struct obd_uuid))
2483 GOTO(out_statfs, rc = -EINVAL);
2485 memcpy(&type, data->ioc_inlbuf1, sizeof(__u32));
2486 if (type & LL_STATFS_LMV)
2487 exp = sbi->ll_md_exp;
2488 else if (type & LL_STATFS_LOV)
2489 exp = sbi->ll_dt_exp;
2491 GOTO(out_statfs, rc = -ENODEV);
2493 rc = obd_iocontrol(IOC_OBD_STATFS, exp, len, buf, NULL);
2495 GOTO(out_statfs, rc);
2497 OBD_FREE_LARGE(buf, len);
2502 * this is normally called in ll_fini_md_op_data(), but sometimes it needs to
2503 * be called early to avoid deadlock.
2505 void ll_unlock_md_op_lsm(struct md_op_data *op_data)
2507 if (op_data->op_mea2_sem) {
2508 up_read(op_data->op_mea2_sem);
2509 op_data->op_mea2_sem = NULL;
2512 if (op_data->op_mea1_sem) {
2513 up_read(op_data->op_mea1_sem);
2514 op_data->op_mea1_sem = NULL;
2518 /* this function prepares md_op_data hint for passing it down to MD stack. */
2519 struct md_op_data *ll_prep_md_op_data(struct md_op_data *op_data,
2520 struct inode *i1, struct inode *i2,
2521 const char *name, size_t namelen,
2522 __u32 mode, __u32 opc, void *data)
2524 LASSERT(i1 != NULL);
2527 /* Do not reuse namelen for something else. */
2529 return ERR_PTR(-EINVAL);
2531 if (namelen > ll_i2sbi(i1)->ll_namelen)
2532 return ERR_PTR(-ENAMETOOLONG);
2534 if (!lu_name_is_valid_2(name, namelen))
2535 return ERR_PTR(-EINVAL);
2538 if (op_data == NULL)
2539 OBD_ALLOC_PTR(op_data);
2541 if (op_data == NULL)
2542 return ERR_PTR(-ENOMEM);
2544 ll_i2gids(op_data->op_suppgids, i1, i2);
2545 op_data->op_fid1 = *ll_inode2fid(i1);
2546 op_data->op_default_stripe_offset = -1;
2548 if (S_ISDIR(i1->i_mode)) {
2549 down_read(&ll_i2info(i1)->lli_lsm_sem);
2550 op_data->op_mea1_sem = &ll_i2info(i1)->lli_lsm_sem;
2551 op_data->op_mea1 = ll_i2info(i1)->lli_lsm_md;
2552 if (opc == LUSTRE_OPC_MKDIR)
2553 op_data->op_default_stripe_offset =
2554 ll_i2info(i1)->lli_def_stripe_offset;
2558 op_data->op_fid2 = *ll_inode2fid(i2);
2559 if (S_ISDIR(i2->i_mode)) {
2561 down_read(&ll_i2info(i2)->lli_lsm_sem);
2562 op_data->op_mea2_sem =
2563 &ll_i2info(i2)->lli_lsm_sem;
2565 op_data->op_mea2 = ll_i2info(i2)->lli_lsm_md;
2568 fid_zero(&op_data->op_fid2);
2571 if (ll_i2sbi(i1)->ll_flags & LL_SBI_64BIT_HASH)
2572 op_data->op_cli_flags |= CLI_HASH64;
2574 if (ll_need_32bit_api(ll_i2sbi(i1)))
2575 op_data->op_cli_flags |= CLI_API32;
2577 op_data->op_name = name;
2578 op_data->op_namelen = namelen;
2579 op_data->op_mode = mode;
2580 op_data->op_mod_time = ktime_get_real_seconds();
2581 op_data->op_fsuid = from_kuid(&init_user_ns, current_fsuid());
2582 op_data->op_fsgid = from_kgid(&init_user_ns, current_fsgid());
2583 op_data->op_cap = cfs_curproc_cap_pack();
2584 op_data->op_mds = 0;
2585 if ((opc == LUSTRE_OPC_CREATE) && (name != NULL) &&
2586 filename_is_volatile(name, namelen, &op_data->op_mds)) {
2587 op_data->op_bias |= MDS_CREATE_VOLATILE;
2589 op_data->op_data = data;
2594 void ll_finish_md_op_data(struct md_op_data *op_data)
2596 ll_unlock_md_op_lsm(op_data);
2597 security_release_secctx(op_data->op_file_secctx,
2598 op_data->op_file_secctx_size);
2599 OBD_FREE_PTR(op_data);
2602 #ifdef HAVE_SUPEROPS_USE_DENTRY
2603 int ll_show_options(struct seq_file *seq, struct dentry *dentry)
2605 int ll_show_options(struct seq_file *seq, struct vfsmount *vfs)
2608 struct ll_sb_info *sbi;
2610 #ifdef HAVE_SUPEROPS_USE_DENTRY
2611 LASSERT((seq != NULL) && (dentry != NULL));
2612 sbi = ll_s2sbi(dentry->d_sb);
2614 LASSERT((seq != NULL) && (vfs != NULL));
2615 sbi = ll_s2sbi(vfs->mnt_sb);
2618 if (sbi->ll_flags & LL_SBI_NOLCK)
2619 seq_puts(seq, ",nolock");
2621 if (sbi->ll_flags & LL_SBI_FLOCK)
2622 seq_puts(seq, ",flock");
2624 if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
2625 seq_puts(seq, ",localflock");
2627 if (sbi->ll_flags & LL_SBI_USER_XATTR)
2628 seq_puts(seq, ",user_xattr");
2630 if (sbi->ll_flags & LL_SBI_LAZYSTATFS)
2631 seq_puts(seq, ",lazystatfs");
2633 if (sbi->ll_flags & LL_SBI_USER_FID2PATH)
2634 seq_puts(seq, ",user_fid2path");
2636 if (sbi->ll_flags & LL_SBI_ALWAYS_PING)
2637 seq_puts(seq, ",always_ping");
2643 * Get obd name by cmd, and copy out to user space
2645 int ll_get_obd_name(struct inode *inode, unsigned int cmd, unsigned long arg)
2647 struct ll_sb_info *sbi = ll_i2sbi(inode);
2648 struct obd_device *obd;
2651 if (cmd == OBD_IOC_GETDTNAME)
2652 obd = class_exp2obd(sbi->ll_dt_exp);
2653 else if (cmd == OBD_IOC_GETMDNAME)
2654 obd = class_exp2obd(sbi->ll_md_exp);
2661 if (copy_to_user((void __user *)arg, obd->obd_name,
2662 strlen(obd->obd_name) + 1))
2669 * Get lustre file system name by \a sbi. If \a buf is provided(non-NULL), the
2670 * fsname will be returned in this buffer; otherwise, a static buffer will be
2671 * used to store the fsname and returned to caller.
2673 char *ll_get_fsname(struct super_block *sb, char *buf, int buflen)
2675 static char fsname_static[MTI_NAME_MAXLEN];
2676 struct lustre_sb_info *lsi = s2lsi(sb);
2681 /* this means the caller wants to use static buffer
2682 * and it doesn't care about race. Usually this is
2683 * in error reporting path */
2684 buf = fsname_static;
2685 buflen = sizeof(fsname_static);
2688 len = strlen(lsi->lsi_lmd->lmd_profile);
2689 ptr = strrchr(lsi->lsi_lmd->lmd_profile, '-');
2690 if (ptr && (strcmp(ptr, "-client") == 0))
2693 if (unlikely(len >= buflen))
2695 strncpy(buf, lsi->lsi_lmd->lmd_profile, len);
2701 static char* ll_d_path(struct dentry *dentry, char *buf, int bufsize)
2708 p.mnt = current->fs->root.mnt;
2710 path = d_path(&p, buf, bufsize);
2715 void ll_dirty_page_discard_warn(struct page *page, int ioret)
2717 char *buf, *path = NULL;
2718 struct dentry *dentry = NULL;
2719 struct inode *inode = page->mapping->host;
2721 /* this can be called inside spin lock so use GFP_ATOMIC. */
2722 buf = (char *)__get_free_page(GFP_ATOMIC);
2724 dentry = d_find_alias(page->mapping->host);
2726 path = ll_d_path(dentry, buf, PAGE_SIZE);
2730 "%s: dirty page discard: %s/fid: "DFID"/%s may get corrupted "
2731 "(rc %d)\n", ll_get_fsname(page->mapping->host->i_sb, NULL, 0),
2732 s2lsi(page->mapping->host->i_sb)->lsi_lmd->lmd_dev,
2733 PFID(ll_inode2fid(inode)),
2734 (path && !IS_ERR(path)) ? path : "", ioret);
2740 free_page((unsigned long)buf);
2743 ssize_t ll_copy_user_md(const struct lov_user_md __user *md,
2744 struct lov_user_md **kbuf)
2746 struct lov_user_md lum;
2750 if (copy_from_user(&lum, md, sizeof(lum)))
2753 lum_size = ll_lov_user_md_size(&lum);
2757 OBD_ALLOC(*kbuf, lum_size);
2761 if (copy_from_user(*kbuf, md, lum_size) != 0) {
2762 OBD_FREE(*kbuf, lum_size);
2770 * Compute llite root squash state after a change of root squash
2771 * configuration setting or add/remove of a lnet nid
2773 void ll_compute_rootsquash_state(struct ll_sb_info *sbi)
2775 struct root_squash_info *squash = &sbi->ll_squash;
2778 struct lnet_process_id id;
2780 /* Update norootsquash flag */
2781 down_write(&squash->rsi_sem);
2782 if (list_empty(&squash->rsi_nosquash_nids))
2783 sbi->ll_flags &= ~LL_SBI_NOROOTSQUASH;
2785 /* Do not apply root squash as soon as one of our NIDs is
2786 * in the nosquash_nids list */
2789 while (LNetGetId(i++, &id) != -ENOENT) {
2790 if (LNET_NETTYP(LNET_NIDNET(id.nid)) == LOLND)
2792 if (cfs_match_nid(id.nid, &squash->rsi_nosquash_nids)) {
2798 sbi->ll_flags |= LL_SBI_NOROOTSQUASH;
2800 sbi->ll_flags &= ~LL_SBI_NOROOTSQUASH;
2802 up_write(&squash->rsi_sem);
2806 * Parse linkea content to extract information about a given hardlink
2808 * \param[in] ldata - Initialized linkea data
2809 * \param[in] linkno - Link identifier
2810 * \param[out] parent_fid - The entry's parent FID
2811 * \param[out] ln - Entry name destination buffer
2813 * \retval 0 on success
2814 * \retval Appropriate negative error code on failure
2816 static int ll_linkea_decode(struct linkea_data *ldata, unsigned int linkno,
2817 struct lu_fid *parent_fid, struct lu_name *ln)
2823 rc = linkea_init_with_rec(ldata);
2827 if (linkno >= ldata->ld_leh->leh_reccount)
2828 /* beyond last link */
2831 linkea_first_entry(ldata);
2832 for (idx = 0; ldata->ld_lee != NULL; idx++) {
2833 linkea_entry_unpack(ldata->ld_lee, &ldata->ld_reclen, ln,
2838 linkea_next_entry(ldata);
2848 * Get parent FID and name of an identified link. Operation is performed for
2849 * a given link number, letting the caller iterate over linkno to list one or
2850 * all links of an entry.
2852 * \param[in] file - File descriptor against which to perform the operation
2853 * \param[in,out] arg - User-filled structure containing the linkno to operate
2854 * on and the available size. It is eventually filled with
2855 * the requested information or left untouched on error
2857 * \retval - 0 on success
2858 * \retval - Appropriate negative error code on failure
2860 int ll_getparent(struct file *file, struct getparent __user *arg)
2862 struct inode *inode = file_inode(file);
2863 struct linkea_data *ldata;
2864 struct lu_buf buf = LU_BUF_NULL;
2866 struct lu_fid parent_fid;
2873 if (!cfs_capable(CFS_CAP_DAC_READ_SEARCH) &&
2874 !(ll_i2sbi(inode)->ll_flags & LL_SBI_USER_FID2PATH))
2877 if (get_user(name_size, &arg->gp_name_size))
2880 if (get_user(linkno, &arg->gp_linkno))
2883 if (name_size > PATH_MAX)
2886 OBD_ALLOC(ldata, sizeof(*ldata));
2890 rc = linkea_data_new(ldata, &buf);
2892 GOTO(ldata_free, rc);
2894 #ifdef HAVE_XATTR_HANDLER_FLAGS
2895 rc = ll_xattr_list(inode, XATTR_NAME_LINK, XATTR_TRUSTED_T, buf.lb_buf,
2896 buf.lb_len, OBD_MD_FLXATTR);
2898 rc = ll_getxattr(file_dentry(file), XATTR_NAME_LINK, buf.lb_buf,
2900 #endif /* HAVE_XATTR_HANDLER_FLAGS */
2904 rc = ll_linkea_decode(ldata, linkno, &parent_fid, &ln);
2908 if (ln.ln_namelen >= name_size)
2909 GOTO(lb_free, rc = -EOVERFLOW);
2911 if (copy_to_user(&arg->gp_fid, &parent_fid, sizeof(arg->gp_fid)))
2912 GOTO(lb_free, rc = -EFAULT);
2914 if (copy_to_user(&arg->gp_name, ln.ln_name, ln.ln_namelen))
2915 GOTO(lb_free, rc = -EFAULT);
2917 if (put_user('\0', arg->gp_name + ln.ln_namelen))
2918 GOTO(lb_free, rc = -EFAULT);
2923 OBD_FREE(ldata, sizeof(*ldata));