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, 2016, 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 #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_max = LL_SA_RPC_DEF;
127 atomic_set(&sbi->ll_sa_total, 0);
128 atomic_set(&sbi->ll_sa_wrong, 0);
129 atomic_set(&sbi->ll_sa_running, 0);
130 atomic_set(&sbi->ll_agl_total, 0);
131 sbi->ll_flags |= LL_SBI_AGL_ENABLED;
132 sbi->ll_flags |= LL_SBI_FAST_READ;
135 sbi->ll_squash.rsi_uid = 0;
136 sbi->ll_squash.rsi_gid = 0;
137 INIT_LIST_HEAD(&sbi->ll_squash.rsi_nosquash_nids);
138 init_rwsem(&sbi->ll_squash.rsi_sem);
143 static void ll_free_sbi(struct super_block *sb)
145 struct ll_sb_info *sbi = ll_s2sbi(sb);
149 if (!list_empty(&sbi->ll_squash.rsi_nosquash_nids))
150 cfs_free_nidlist(&sbi->ll_squash.rsi_nosquash_nids);
151 if (sbi->ll_cache != NULL) {
152 cl_cache_decref(sbi->ll_cache);
153 sbi->ll_cache = NULL;
155 OBD_FREE(sbi, sizeof(*sbi));
160 static inline int obd_connect_has_secctx(struct obd_connect_data *data)
162 return data->ocd_connect_flags & OBD_CONNECT_FLAGS2 &&
163 data->ocd_connect_flags2 & OBD_CONNECT2_FILE_SECCTX;
166 static int client_common_fill_super(struct super_block *sb, char *md, char *dt,
167 struct vfsmount *mnt)
169 struct inode *root = NULL;
170 struct ll_sb_info *sbi = ll_s2sbi(sb);
171 struct obd_device *obd;
172 struct obd_statfs *osfs = NULL;
173 struct ptlrpc_request *request = NULL;
174 struct obd_connect_data *data = NULL;
175 struct obd_uuid *uuid;
176 struct md_op_data *op_data;
177 struct lustre_md lmd;
179 int size, err, checksum;
182 obd = class_name2obd(md);
184 CERROR("MD %s: not setup or attached\n", md);
198 /* indicate MDT features supported by this client */
199 data->ocd_connect_flags = OBD_CONNECT_IBITS | OBD_CONNECT_NODEVOH |
200 OBD_CONNECT_ATTRFID |
201 OBD_CONNECT_VERSION | OBD_CONNECT_BRW_SIZE |
202 OBD_CONNECT_MDS_CAPA | OBD_CONNECT_OSS_CAPA |
203 OBD_CONNECT_CANCELSET | OBD_CONNECT_FID |
204 OBD_CONNECT_AT | OBD_CONNECT_LOV_V3 |
205 OBD_CONNECT_VBR | OBD_CONNECT_FULL20 |
206 OBD_CONNECT_64BITHASH |
207 OBD_CONNECT_EINPROGRESS |
208 OBD_CONNECT_JOBSTATS | OBD_CONNECT_LVB_TYPE |
209 OBD_CONNECT_LAYOUTLOCK | OBD_CONNECT_PINGLESS|
210 OBD_CONNECT_MAX_EASIZE |
211 OBD_CONNECT_FLOCK_DEAD |
212 OBD_CONNECT_DISP_STRIPE | OBD_CONNECT_LFSCK |
213 OBD_CONNECT_OPEN_BY_FID |
214 OBD_CONNECT_DIR_STRIPE |
215 OBD_CONNECT_BULK_MBITS |
216 OBD_CONNECT_SUBTREE |
217 OBD_CONNECT_FLAGS2 | OBD_CONNECT_MULTIMODRPCS;
219 data->ocd_connect_flags2 = 0;
221 #ifdef HAVE_LRU_RESIZE_SUPPORT
222 if (sbi->ll_flags & LL_SBI_LRU_RESIZE)
223 data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
225 #ifdef CONFIG_FS_POSIX_ACL
226 data->ocd_connect_flags |= OBD_CONNECT_ACL | OBD_CONNECT_UMASK |
227 OBD_CONNECT_LARGE_ACL;
230 if (OBD_FAIL_CHECK(OBD_FAIL_MDC_LIGHTWEIGHT))
231 /* flag mdc connection as lightweight, only used for test
232 * purpose, use with care */
233 data->ocd_connect_flags |= OBD_CONNECT_LIGHTWEIGHT;
235 data->ocd_ibits_known = MDS_INODELOCK_FULL;
236 data->ocd_version = LUSTRE_VERSION_CODE;
238 if (sb->s_flags & MS_RDONLY)
239 data->ocd_connect_flags |= OBD_CONNECT_RDONLY;
240 if (sbi->ll_flags & LL_SBI_USER_XATTR)
241 data->ocd_connect_flags |= OBD_CONNECT_XATTR;
244 /* Setting this indicates we correctly support S_NOSEC (See kernel
245 * commit 9e1f1de02c2275d7172e18dc4e7c2065777611bf)
247 sb->s_flags |= MS_NOSEC;
250 if (sbi->ll_flags & LL_SBI_FLOCK)
251 sbi->ll_fop = &ll_file_operations_flock;
252 else if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
253 sbi->ll_fop = &ll_file_operations;
255 sbi->ll_fop = &ll_file_operations_noflock;
257 /* always ping even if server suppress_pings */
258 if (sbi->ll_flags & LL_SBI_ALWAYS_PING)
259 data->ocd_connect_flags &= ~OBD_CONNECT_PINGLESS;
261 #if defined(HAVE_SECURITY_DENTRY_INIT_SECURITY) && defined(CONFIG_SECURITY)
262 data->ocd_connect_flags2 |= OBD_CONNECT2_FILE_SECCTX;
263 #endif /* HAVE_SECURITY_DENTRY_INIT_SECURITY */
265 data->ocd_brw_size = MD_MAX_BRW_SIZE;
267 err = obd_connect(NULL, &sbi->ll_md_exp, obd, &sbi->ll_sb_uuid, data, NULL);
269 LCONSOLE_ERROR_MSG(0x14f, "An MDT (md %s) is performing "
270 "recovery, of which this client is not a "
271 "part. Please wait for recovery to complete,"
272 " abort, or time out.\n", md);
275 CERROR("cannot connect to %s: rc = %d\n", md, err);
279 sbi->ll_md_exp->exp_connect_data = *data;
281 err = obd_fid_init(sbi->ll_md_exp->exp_obd, sbi->ll_md_exp,
282 LUSTRE_SEQ_METADATA);
284 CERROR("%s: Can't init metadata layer FID infrastructure, "
285 "rc = %d\n", sbi->ll_md_exp->exp_obd->obd_name, err);
289 /* For mount, we only need fs info from MDT0, and also in DNE, it
290 * can make sure the client can be mounted as long as MDT0 is
292 err = obd_statfs(NULL, sbi->ll_md_exp, osfs,
293 cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS),
294 OBD_STATFS_FOR_MDT0);
296 GOTO(out_md_fid, err);
298 /* This needs to be after statfs to ensure connect has finished.
299 * Note that "data" does NOT contain the valid connect reply.
300 * If connecting to a 1.8 server there will be no LMV device, so
301 * we can access the MDC export directly and exp_connect_flags will
302 * be non-zero, but if accessing an upgraded 2.1 server it will
303 * have the correct flags filled in.
304 * XXX: fill in the LMV exp_connect_flags from MDC(s). */
305 valid = exp_connect_flags(sbi->ll_md_exp) & CLIENT_CONNECT_MDT_REQD;
306 if (exp_connect_flags(sbi->ll_md_exp) != 0 &&
307 valid != CLIENT_CONNECT_MDT_REQD) {
310 OBD_ALLOC_WAIT(buf, PAGE_SIZE);
311 obd_connect_flags2str(buf, PAGE_SIZE,
312 valid ^ CLIENT_CONNECT_MDT_REQD, 0, ",");
313 LCONSOLE_ERROR_MSG(0x170, "Server %s does not support "
314 "feature(s) needed for correct operation "
315 "of this client (%s). Please upgrade "
316 "server or downgrade client.\n",
317 sbi->ll_md_exp->exp_obd->obd_name, buf);
318 OBD_FREE(buf, PAGE_SIZE);
319 GOTO(out_md_fid, err = -EPROTO);
322 size = sizeof(*data);
323 err = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_CONN_DATA),
324 KEY_CONN_DATA, &size, data);
326 CERROR("%s: Get connect data failed: rc = %d\n",
327 sbi->ll_md_exp->exp_obd->obd_name, err);
328 GOTO(out_md_fid, err);
331 LASSERT(osfs->os_bsize);
332 sb->s_blocksize = osfs->os_bsize;
333 sb->s_blocksize_bits = log2(osfs->os_bsize);
334 sb->s_magic = LL_SUPER_MAGIC;
335 sb->s_maxbytes = MAX_LFS_FILESIZE;
336 sbi->ll_namelen = osfs->os_namelen;
337 sbi->ll_mnt.mnt = current->fs->root.mnt;
339 if ((sbi->ll_flags & LL_SBI_USER_XATTR) &&
340 !(data->ocd_connect_flags & OBD_CONNECT_XATTR)) {
341 LCONSOLE_INFO("Disabling user_xattr feature because "
342 "it is not supported on the server\n");
343 sbi->ll_flags &= ~LL_SBI_USER_XATTR;
346 if (data->ocd_connect_flags & OBD_CONNECT_ACL) {
348 sb->s_flags |= MS_POSIXACL;
350 sbi->ll_flags |= LL_SBI_ACL;
352 LCONSOLE_INFO("client wants to enable acl, but mdt not!\n");
354 sb->s_flags &= ~MS_POSIXACL;
356 sbi->ll_flags &= ~LL_SBI_ACL;
359 if (data->ocd_connect_flags & OBD_CONNECT_64BITHASH)
360 sbi->ll_flags |= LL_SBI_64BIT_HASH;
362 if (data->ocd_connect_flags & OBD_CONNECT_LAYOUTLOCK)
363 sbi->ll_flags |= LL_SBI_LAYOUT_LOCK;
365 if (obd_connect_has_secctx(data))
366 sbi->ll_flags |= LL_SBI_FILE_SECCTX;
368 if (data->ocd_ibits_known & MDS_INODELOCK_XATTR) {
369 if (!(data->ocd_connect_flags & OBD_CONNECT_MAX_EASIZE)) {
370 LCONSOLE_INFO("%s: disabling xattr cache due to "
371 "unknown maximum xattr size.\n", dt);
372 } else if (!sbi->ll_xattr_cache_set) {
373 /* If xattr_cache is already set (no matter 0 or 1)
374 * during processing llog, it won't be enabled here. */
375 sbi->ll_flags |= LL_SBI_XATTR_CACHE;
376 sbi->ll_xattr_cache_enabled = 1;
380 obd = class_name2obd(dt);
382 CERROR("DT %s: not setup or attached\n", dt);
383 GOTO(out_md_fid, err = -ENODEV);
386 /* pass client page size via ocd_grant_blkbits, the server should report
387 * back its backend blocksize for grant calculation purpose */
388 data->ocd_grant_blkbits = PAGE_SHIFT;
390 /* indicate OST features supported by this client */
391 data->ocd_connect_flags = OBD_CONNECT_GRANT | OBD_CONNECT_VERSION |
392 OBD_CONNECT_REQPORTAL | OBD_CONNECT_BRW_SIZE |
393 OBD_CONNECT_CANCELSET | OBD_CONNECT_FID |
394 OBD_CONNECT_SRVLOCK | OBD_CONNECT_TRUNCLOCK|
395 OBD_CONNECT_AT | OBD_CONNECT_OSS_CAPA |
396 OBD_CONNECT_VBR | OBD_CONNECT_FULL20 |
397 OBD_CONNECT_64BITHASH | OBD_CONNECT_MAXBYTES |
398 OBD_CONNECT_EINPROGRESS |
399 OBD_CONNECT_JOBSTATS | OBD_CONNECT_LVB_TYPE |
400 OBD_CONNECT_LAYOUTLOCK |
401 OBD_CONNECT_PINGLESS | OBD_CONNECT_LFSCK |
402 OBD_CONNECT_BULK_MBITS | OBD_CONNECT_SHORTIO |
405 /* The client currently advertises support for OBD_CONNECT_LOCKAHEAD_OLD so it
406 * can interoperate with an older version of lockahead which was released prior
407 * to landing in master. This support will be dropped when 2.13 development
408 * starts. At the point, we should not just drop the connect flag (below), we
409 * should also remove the support in the code.
411 * Removing it means a few things:
412 * 1. Remove this section here
413 * 2. Remove CEF_NONBLOCK in ll_file_lockahead()
414 * 3. Remove function exp_connect_lockahead_old
415 * 4. Remove LDLM_FL_LOCKAHEAD_OLD_RESERVED in lustre_dlm_flags.h
417 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 12, 50, 0)
418 data->ocd_connect_flags |= OBD_CONNECT_LOCKAHEAD_OLD;
421 data->ocd_connect_flags2 = OBD_CONNECT2_LOCKAHEAD;
423 if (!OBD_FAIL_CHECK(OBD_FAIL_OSC_CONNECT_GRANT_PARAM))
424 data->ocd_connect_flags |= OBD_CONNECT_GRANT_PARAM;
426 /* OBD_CONNECT_CKSUM should always be set, even if checksums are
427 * disabled by default, because it can still be enabled on the
428 * fly via /proc. As a consequence, we still need to come to an
429 * agreement on the supported algorithms at connect time */
430 data->ocd_connect_flags |= OBD_CONNECT_CKSUM;
432 if (OBD_FAIL_CHECK(OBD_FAIL_OSC_CKSUM_ADLER_ONLY))
433 data->ocd_cksum_types = OBD_CKSUM_ADLER;
435 data->ocd_cksum_types = cksum_types_supported_client();
437 #ifdef HAVE_LRU_RESIZE_SUPPORT
438 data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
440 /* always ping even if server suppress_pings */
441 if (sbi->ll_flags & LL_SBI_ALWAYS_PING)
442 data->ocd_connect_flags &= ~OBD_CONNECT_PINGLESS;
444 CDEBUG(D_RPCTRACE, "ocd_connect_flags: %#llx ocd_version: %d "
445 "ocd_grant: %d\n", data->ocd_connect_flags,
446 data->ocd_version, data->ocd_grant);
448 obd->obd_upcall.onu_owner = &sbi->ll_lco;
449 obd->obd_upcall.onu_upcall = cl_ocd_update;
451 data->ocd_brw_size = DT_MAX_BRW_SIZE;
453 err = obd_connect(NULL, &sbi->ll_dt_exp, obd, &sbi->ll_sb_uuid, data,
456 LCONSOLE_ERROR_MSG(0x150, "An OST (dt %s) is performing "
457 "recovery, of which this client is not a "
458 "part. Please wait for recovery to "
459 "complete, abort, or time out.\n", dt);
462 CERROR("%s: Cannot connect to %s: rc = %d\n",
463 sbi->ll_dt_exp->exp_obd->obd_name, dt, err);
467 sbi->ll_dt_exp->exp_connect_data = *data;
469 /* Don't change value if it was specified in the config log */
470 if (sbi->ll_ra_info.ra_max_read_ahead_whole_pages == -1)
471 sbi->ll_ra_info.ra_max_read_ahead_whole_pages =
472 max_t(unsigned long, SBI_DEFAULT_READAHEAD_WHOLE_MAX,
473 (data->ocd_brw_size >> PAGE_SHIFT));
475 err = obd_fid_init(sbi->ll_dt_exp->exp_obd, sbi->ll_dt_exp,
476 LUSTRE_SEQ_METADATA);
478 CERROR("%s: Can't init data layer FID infrastructure, "
479 "rc = %d\n", sbi->ll_dt_exp->exp_obd->obd_name, err);
483 mutex_lock(&sbi->ll_lco.lco_lock);
484 sbi->ll_lco.lco_flags = data->ocd_connect_flags;
485 sbi->ll_lco.lco_md_exp = sbi->ll_md_exp;
486 sbi->ll_lco.lco_dt_exp = sbi->ll_dt_exp;
487 mutex_unlock(&sbi->ll_lco.lco_lock);
489 fid_zero(&sbi->ll_root_fid);
490 err = md_get_root(sbi->ll_md_exp, get_mount_fileset(sb),
493 CERROR("cannot mds_connect: rc = %d\n", err);
494 GOTO(out_lock_cn_cb, err);
496 if (!fid_is_sane(&sbi->ll_root_fid)) {
497 CERROR("%s: Invalid root fid "DFID" during mount\n",
498 sbi->ll_md_exp->exp_obd->obd_name,
499 PFID(&sbi->ll_root_fid));
500 GOTO(out_lock_cn_cb, err = -EINVAL);
502 CDEBUG(D_SUPER, "rootfid "DFID"\n", PFID(&sbi->ll_root_fid));
504 sb->s_op = &lustre_super_operations;
505 #ifdef HAVE_XATTR_HANDLER_FLAGS
506 sb->s_xattr = ll_xattr_handlers;
508 #if THREAD_SIZE >= 8192 /*b=17630*/
509 sb->s_export_op = &lustre_export_operations;
513 * XXX: move this to after cbd setup? */
514 valid = OBD_MD_FLGETATTR | OBD_MD_FLBLOCKS | OBD_MD_FLMODEASIZE;
515 if (sbi->ll_flags & LL_SBI_ACL)
516 valid |= OBD_MD_FLACL;
518 OBD_ALLOC_PTR(op_data);
520 GOTO(out_lock_cn_cb, err = -ENOMEM);
522 op_data->op_fid1 = sbi->ll_root_fid;
523 op_data->op_mode = 0;
524 op_data->op_valid = valid;
526 err = md_getattr(sbi->ll_md_exp, op_data, &request);
528 OBD_FREE_PTR(op_data);
530 CERROR("%s: md_getattr failed for root: rc = %d\n",
531 sbi->ll_md_exp->exp_obd->obd_name, err);
532 GOTO(out_lock_cn_cb, err);
535 err = md_get_lustre_md(sbi->ll_md_exp, request, sbi->ll_dt_exp,
536 sbi->ll_md_exp, &lmd);
538 CERROR("failed to understand root inode md: rc = %d\n", err);
539 ptlrpc_req_finished(request);
540 GOTO(out_lock_cn_cb, err);
543 LASSERT(fid_is_sane(&sbi->ll_root_fid));
544 root = ll_iget(sb, cl_fid_build_ino(&sbi->ll_root_fid,
545 sbi->ll_flags & LL_SBI_32BIT_API),
547 md_free_lustre_md(sbi->ll_md_exp, &lmd);
548 ptlrpc_req_finished(request);
551 #ifdef CONFIG_FS_POSIX_ACL
553 posix_acl_release(lmd.posix_acl);
554 lmd.posix_acl = NULL;
557 err = IS_ERR(root) ? PTR_ERR(root) : -EBADF;
559 CERROR("lustre_lite: bad iget4 for root\n");
563 checksum = sbi->ll_flags & LL_SBI_CHECKSUM;
564 err = obd_set_info_async(NULL, sbi->ll_dt_exp, sizeof(KEY_CHECKSUM),
565 KEY_CHECKSUM, sizeof(checksum), &checksum,
568 CERROR("%s: Set checksum failed: rc = %d\n",
569 sbi->ll_dt_exp->exp_obd->obd_name, err);
574 err = obd_set_info_async(NULL, sbi->ll_dt_exp, sizeof(KEY_CACHE_SET),
575 KEY_CACHE_SET, sizeof(*sbi->ll_cache),
576 sbi->ll_cache, NULL);
578 CERROR("%s: Set cache_set failed: rc = %d\n",
579 sbi->ll_dt_exp->exp_obd->obd_name, err);
583 sb->s_root = d_make_root(root);
584 if (sb->s_root == NULL) {
585 CERROR("%s: can't make root dentry\n",
586 ll_get_fsname(sb, NULL, 0));
587 GOTO(out_root, err = -ENOMEM);
589 #ifdef HAVE_DCACHE_LOCK
590 sb->s_root->d_op = &ll_d_ops;
593 sbi->ll_sdev_orig = sb->s_dev;
595 /* We set sb->s_dev equal on all lustre clients in order to support
596 * NFS export clustering. NFSD requires that the FSID be the same
598 /* s_dev is also used in lt_compare() to compare two fs, but that is
599 * only a node-local comparison. */
600 uuid = obd_get_uuid(sbi->ll_md_exp);
602 sb->s_dev = get_uuid2int(uuid->uuid, strlen(uuid->uuid));
609 if (sbi->ll_proc_root != NULL) {
610 err = lprocfs_ll_register_obd(sb, dt);
612 CERROR("%s: could not register %s in llite: rc = %d\n",
613 dt, ll_get_fsname(sb, NULL, 0), err);
616 err = lprocfs_ll_register_obd(sb, md);
618 CERROR("%s: could not register %s in llite: rc = %d\n",
619 md, ll_get_fsname(sb, NULL, 0), err);
629 obd_fid_fini(sbi->ll_dt_exp->exp_obd);
631 obd_disconnect(sbi->ll_dt_exp);
632 sbi->ll_dt_exp = NULL;
634 obd_fid_fini(sbi->ll_md_exp->exp_obd);
636 obd_disconnect(sbi->ll_md_exp);
637 sbi->ll_md_exp = NULL;
646 int ll_get_max_mdsize(struct ll_sb_info *sbi, int *lmmsize)
650 size = sizeof(*lmmsize);
651 rc = obd_get_info(NULL, sbi->ll_dt_exp, sizeof(KEY_MAX_EASIZE),
652 KEY_MAX_EASIZE, &size, lmmsize);
654 CERROR("%s: cannot get max LOV EA size: rc = %d\n",
655 sbi->ll_dt_exp->exp_obd->obd_name, rc);
660 rc = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_MAX_EASIZE),
661 KEY_MAX_EASIZE, &size, lmmsize);
663 CERROR("Get max mdsize error rc %d\n", rc);
669 * Get the value of the default_easize parameter.
671 * \see client_obd::cl_default_mds_easize
673 * \param[in] sbi superblock info for this filesystem
674 * \param[out] lmmsize pointer to storage location for value
676 * \retval 0 on success
677 * \retval negative negated errno on failure
679 int ll_get_default_mdsize(struct ll_sb_info *sbi, int *lmmsize)
684 rc = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_DEFAULT_EASIZE),
685 KEY_DEFAULT_EASIZE, &size, lmmsize);
687 CERROR("Get default mdsize error rc %d\n", rc);
693 * Set the default_easize parameter to the given value.
695 * \see client_obd::cl_default_mds_easize
697 * \param[in] sbi superblock info for this filesystem
698 * \param[in] lmmsize the size to set
700 * \retval 0 on success
701 * \retval negative negated errno on failure
703 int ll_set_default_mdsize(struct ll_sb_info *sbi, int lmmsize)
707 if (lmmsize < sizeof(struct lov_mds_md) ||
708 lmmsize > OBD_MAX_DEFAULT_EA_SIZE)
711 rc = obd_set_info_async(NULL, sbi->ll_md_exp,
712 sizeof(KEY_DEFAULT_EASIZE), KEY_DEFAULT_EASIZE,
713 sizeof(int), &lmmsize, NULL);
718 static void client_common_put_super(struct super_block *sb)
720 struct ll_sb_info *sbi = ll_s2sbi(sb);
725 obd_fid_fini(sbi->ll_dt_exp->exp_obd);
726 obd_disconnect(sbi->ll_dt_exp);
727 sbi->ll_dt_exp = NULL;
729 lprocfs_ll_unregister_mountpoint(sbi);
731 obd_fid_fini(sbi->ll_md_exp->exp_obd);
732 obd_disconnect(sbi->ll_md_exp);
733 sbi->ll_md_exp = NULL;
738 void ll_kill_super(struct super_block *sb)
740 struct ll_sb_info *sbi;
744 if (!(sb->s_flags & MS_ACTIVE))
748 /* we need restore s_dev from changed for clustred NFS before put_super
749 * because new kernels have cached s_dev and change sb->s_dev in
750 * put_super not affected real removing devices */
752 sb->s_dev = sbi->ll_sdev_orig;
753 sbi->ll_umounting = 1;
755 /* wait running statahead threads to quit */
756 while (atomic_read(&sbi->ll_sa_running) > 0) {
757 set_current_state(TASK_UNINTERRUPTIBLE);
758 schedule_timeout(msecs_to_jiffies(MSEC_PER_SEC >> 3));
765 static inline int ll_set_opt(const char *opt, char *data, int fl)
767 if (strncmp(opt, data, strlen(opt)) != 0)
773 /* non-client-specific mount options are parsed in lmd_parse */
774 static int ll_options(char *options, int *flags)
777 char *s1 = options, *s2;
783 CDEBUG(D_CONFIG, "Parsing opts %s\n", options);
786 CDEBUG(D_SUPER, "next opt=%s\n", s1);
787 tmp = ll_set_opt("nolock", s1, LL_SBI_NOLCK);
792 tmp = ll_set_opt("flock", s1, LL_SBI_FLOCK);
797 tmp = ll_set_opt("localflock", s1, LL_SBI_LOCALFLOCK);
802 tmp = ll_set_opt("noflock", s1, LL_SBI_FLOCK|LL_SBI_LOCALFLOCK);
807 tmp = ll_set_opt("user_xattr", s1, LL_SBI_USER_XATTR);
812 tmp = ll_set_opt("nouser_xattr", s1, LL_SBI_USER_XATTR);
817 tmp = ll_set_opt("context", s1, 1);
820 tmp = ll_set_opt("fscontext", s1, 1);
823 tmp = ll_set_opt("defcontext", s1, 1);
826 tmp = ll_set_opt("rootcontext", s1, 1);
829 tmp = ll_set_opt("user_fid2path", s1, LL_SBI_USER_FID2PATH);
834 tmp = ll_set_opt("nouser_fid2path", s1, LL_SBI_USER_FID2PATH);
840 tmp = ll_set_opt("checksum", s1, LL_SBI_CHECKSUM);
845 tmp = ll_set_opt("nochecksum", s1, LL_SBI_CHECKSUM);
850 tmp = ll_set_opt("lruresize", s1, LL_SBI_LRU_RESIZE);
855 tmp = ll_set_opt("nolruresize", s1, LL_SBI_LRU_RESIZE);
860 tmp = ll_set_opt("lazystatfs", s1, LL_SBI_LAZYSTATFS);
865 tmp = ll_set_opt("nolazystatfs", s1, LL_SBI_LAZYSTATFS);
870 tmp = ll_set_opt("32bitapi", s1, LL_SBI_32BIT_API);
875 tmp = ll_set_opt("verbose", s1, LL_SBI_VERBOSE);
880 tmp = ll_set_opt("noverbose", s1, LL_SBI_VERBOSE);
885 tmp = ll_set_opt("always_ping", s1, LL_SBI_ALWAYS_PING);
890 LCONSOLE_ERROR_MSG(0x152, "Unknown option '%s', won't mount.\n",
896 s2 = strchr(s1, ',');
904 void ll_lli_init(struct ll_inode_info *lli)
906 lli->lli_inode_magic = LLI_INODE_MAGIC;
908 spin_lock_init(&lli->lli_lock);
909 lli->lli_posix_acl = NULL;
910 /* Do not set lli_fid, it has been initialized already. */
911 fid_zero(&lli->lli_pfid);
912 lli->lli_mds_read_och = NULL;
913 lli->lli_mds_write_och = NULL;
914 lli->lli_mds_exec_och = NULL;
915 lli->lli_open_fd_read_count = 0;
916 lli->lli_open_fd_write_count = 0;
917 lli->lli_open_fd_exec_count = 0;
918 mutex_init(&lli->lli_och_mutex);
919 spin_lock_init(&lli->lli_agl_lock);
920 spin_lock_init(&lli->lli_layout_lock);
921 ll_layout_version_set(lli, CL_LAYOUT_GEN_NONE);
922 lli->lli_clob = NULL;
924 init_rwsem(&lli->lli_xattrs_list_rwsem);
925 mutex_init(&lli->lli_xattrs_enq_lock);
927 LASSERT(lli->lli_vfs_inode.i_mode != 0);
928 if (S_ISDIR(lli->lli_vfs_inode.i_mode)) {
929 mutex_init(&lli->lli_readdir_mutex);
930 lli->lli_opendir_key = NULL;
932 spin_lock_init(&lli->lli_sa_lock);
933 lli->lli_opendir_pid = 0;
934 lli->lli_sa_enabled = 0;
935 lli->lli_def_stripe_offset = -1;
937 mutex_init(&lli->lli_size_mutex);
938 lli->lli_symlink_name = NULL;
939 init_rwsem(&lli->lli_trunc_sem);
940 range_lock_tree_init(&lli->lli_write_tree);
941 init_rwsem(&lli->lli_glimpse_sem);
942 lli->lli_glimpse_time = 0;
943 INIT_LIST_HEAD(&lli->lli_agl_list);
944 lli->lli_agl_index = 0;
945 lli->lli_async_rc = 0;
947 mutex_init(&lli->lli_layout_mutex);
948 memset(lli->lli_jobid, 0, LUSTRE_JOBID_SIZE);
951 #ifndef HAVE_SUPER_SETUP_BDI_NAME
953 #define LSI_BDI_INITIALIZED 0x00400000
955 #ifndef HAVE_BDI_CAP_MAP_COPY
956 # define BDI_CAP_MAP_COPY 0
959 #define MAX_STRING_SIZE 128
961 static int super_setup_bdi_name(struct super_block *sb, char *fmt, ...)
963 struct lustre_sb_info *lsi = s2lsi(sb);
964 char buf[MAX_STRING_SIZE];
968 err = bdi_init(&lsi->lsi_bdi);
972 lsi->lsi_flags |= LSI_BDI_INITIALIZED;
973 lsi->lsi_bdi.capabilities = BDI_CAP_MAP_COPY;
974 lsi->lsi_bdi.name = "lustre";
976 vsnprintf(buf, MAX_STRING_SIZE, fmt, args);
978 err = bdi_register(&lsi->lsi_bdi, NULL, "%s", buf);
981 sb->s_bdi = &lsi->lsi_bdi;
985 #endif /* !HAVE_SUPER_SETUP_BDI_NAME */
987 int ll_fill_super(struct super_block *sb, struct vfsmount *mnt)
989 struct lustre_profile *lprof = NULL;
990 struct lustre_sb_info *lsi = s2lsi(sb);
991 struct ll_sb_info *sbi;
992 char *dt = NULL, *md = NULL;
993 char *profilenm = get_profile_name(sb);
994 struct config_llog_instance *cfg;
995 /* %p for void* in printf needs 16+2 characters: 0xffffffffffffffff */
996 const int instlen = sizeof(cfg->cfg_instance) * 2 + 2;
1002 CDEBUG(D_VFSTRACE, "VFS Op: sb %p\n", sb);
1008 try_module_get(THIS_MODULE);
1010 /* client additional sb info */
1011 lsi->lsi_llsbi = sbi = ll_init_sbi();
1013 module_put(THIS_MODULE);
1018 err = ll_options(lsi->lsi_lmd->lmd_opts, &sbi->ll_flags);
1020 GOTO(out_free, err);
1022 err = super_setup_bdi_name(sb, "lustre-%p", sb);
1024 GOTO(out_free, err);
1026 #ifndef HAVE_DCACHE_LOCK
1027 /* kernel >= 2.6.38 store dentry operations in sb->s_d_op. */
1028 sb->s_d_op = &ll_d_ops;
1031 /* Call lprocfs_ll_register_mountpoint() before lustre_process_log()
1032 * so that "llite.*.*" params can be processed correctly. */
1033 if (proc_lustre_fs_root != NULL) {
1034 err = lprocfs_ll_register_mountpoint(proc_lustre_fs_root, sb);
1036 CERROR("%s: could not register mountpoint in llite: "
1037 "rc = %d\n", ll_get_fsname(sb, NULL, 0), err);
1042 /* Generate a string unique to this super, in case some joker tries
1043 to mount the same fs at two mount points.
1044 Use the address of the super itself.*/
1045 cfg->cfg_instance = sb;
1046 cfg->cfg_uuid = lsi->lsi_llsbi->ll_sb_uuid;
1047 cfg->cfg_callback = class_config_llog_handler;
1048 cfg->cfg_sub_clds = CONFIG_SUB_CLIENT;
1049 /* set up client obds */
1050 err = lustre_process_log(sb, profilenm, cfg);
1052 GOTO(out_proc, err);
1054 /* Profile set with LCFG_MOUNTOPT so we can find our mdc and osc obds */
1055 lprof = class_get_profile(profilenm);
1056 if (lprof == NULL) {
1057 LCONSOLE_ERROR_MSG(0x156, "The client profile '%s' could not be"
1058 " read from the MGS. Does that filesystem "
1059 "exist?\n", profilenm);
1060 GOTO(out_proc, err = -EINVAL);
1062 CDEBUG(D_CONFIG, "Found profile %s: mdc=%s osc=%s\n", profilenm,
1063 lprof->lp_md, lprof->lp_dt);
1065 dt_len = strlen(lprof->lp_dt) + instlen + 2;
1066 OBD_ALLOC(dt, dt_len);
1068 GOTO(out_proc, err = -ENOMEM);
1069 snprintf(dt, dt_len - 1, "%s-%p", lprof->lp_dt, cfg->cfg_instance);
1071 md_len = strlen(lprof->lp_md) + instlen + 2;
1072 OBD_ALLOC(md, md_len);
1074 GOTO(out_proc, err = -ENOMEM);
1075 snprintf(md, md_len - 1, "%s-%p", lprof->lp_md, cfg->cfg_instance);
1077 /* connections, registrations, sb setup */
1078 err = client_common_fill_super(sb, md, dt, mnt);
1080 GOTO(out_proc, err);
1082 sbi->ll_client_common_fill_super_succeeded = 1;
1086 lprocfs_ll_unregister_mountpoint(sbi);
1089 OBD_FREE(md, md_len);
1091 OBD_FREE(dt, dt_len);
1093 class_put_profile(lprof);
1096 else if (sbi->ll_flags & LL_SBI_VERBOSE)
1097 LCONSOLE_WARN("Mounted %s\n", profilenm);
1101 } /* ll_fill_super */
1103 void ll_put_super(struct super_block *sb)
1105 struct config_llog_instance cfg, params_cfg;
1106 struct obd_device *obd;
1107 struct lustre_sb_info *lsi = s2lsi(sb);
1108 struct ll_sb_info *sbi = ll_s2sbi(sb);
1109 char *profilenm = get_profile_name(sb);
1111 int next, force = 1, rc = 0;
1114 CDEBUG(D_VFSTRACE, "VFS Op: sb %p - %s\n", sb, profilenm);
1116 cfg.cfg_instance = sb;
1117 lustre_end_log(sb, profilenm, &cfg);
1119 params_cfg.cfg_instance = sb;
1120 lustre_end_log(sb, PARAMS_FILENAME, ¶ms_cfg);
1122 if (sbi->ll_md_exp) {
1123 obd = class_exp2obd(sbi->ll_md_exp);
1125 force = obd->obd_force;
1128 /* Wait for unstable pages to be committed to stable storage */
1130 struct l_wait_info lwi = LWI_INTR(LWI_ON_SIGNAL_NOOP, NULL);
1131 rc = l_wait_event(sbi->ll_cache->ccc_unstable_waitq,
1132 atomic_long_read(&sbi->ll_cache->ccc_unstable_nr) == 0,
1136 ccc_count = atomic_long_read(&sbi->ll_cache->ccc_unstable_nr);
1137 if (force == 0 && rc != -EINTR)
1138 LASSERTF(ccc_count == 0, "count: %li\n", ccc_count);
1141 /* We need to set force before the lov_disconnect in
1142 lustre_common_put_super, since l_d cleans up osc's as well. */
1145 while ((obd = class_devices_in_group(&sbi->ll_sb_uuid,
1147 obd->obd_force = force;
1151 if (sbi->ll_client_common_fill_super_succeeded) {
1152 /* Only if client_common_fill_super succeeded */
1153 client_common_put_super(sb);
1157 while ((obd = class_devices_in_group(&sbi->ll_sb_uuid, &next)) !=NULL) {
1158 class_manual_cleanup(obd);
1161 if (sbi->ll_flags & LL_SBI_VERBOSE)
1162 LCONSOLE_WARN("Unmounted %s\n", profilenm ? profilenm : "");
1165 class_del_profile(profilenm);
1167 #ifndef HAVE_SUPER_SETUP_BDI_NAME
1168 if (lsi->lsi_flags & LSI_BDI_INITIALIZED) {
1169 bdi_destroy(&lsi->lsi_bdi);
1170 lsi->lsi_flags &= ~LSI_BDI_INITIALIZED;
1175 lsi->lsi_llsbi = NULL;
1177 lustre_common_put_super(sb);
1179 cl_env_cache_purge(~0);
1181 module_put(THIS_MODULE);
1184 } /* client_put_super */
1186 struct inode *ll_inode_from_resource_lock(struct ldlm_lock *lock)
1188 struct inode *inode = NULL;
1190 /* NOTE: we depend on atomic igrab() -bzzz */
1191 lock_res_and_lock(lock);
1192 if (lock->l_resource->lr_lvb_inode) {
1193 struct ll_inode_info * lli;
1194 lli = ll_i2info(lock->l_resource->lr_lvb_inode);
1195 if (lli->lli_inode_magic == LLI_INODE_MAGIC) {
1196 inode = igrab(lock->l_resource->lr_lvb_inode);
1198 inode = lock->l_resource->lr_lvb_inode;
1199 LDLM_DEBUG_LIMIT(inode->i_state & I_FREEING ? D_INFO :
1200 D_WARNING, lock, "lr_lvb_inode %p is "
1201 "bogus: magic %08x",
1202 lock->l_resource->lr_lvb_inode,
1203 lli->lli_inode_magic);
1207 unlock_res_and_lock(lock);
1211 void ll_dir_clear_lsm_md(struct inode *inode)
1213 struct ll_inode_info *lli = ll_i2info(inode);
1215 LASSERT(S_ISDIR(inode->i_mode));
1217 if (lli->lli_lsm_md != NULL) {
1218 lmv_free_memmd(lli->lli_lsm_md);
1219 lli->lli_lsm_md = NULL;
1223 static struct inode *ll_iget_anon_dir(struct super_block *sb,
1224 const struct lu_fid *fid,
1225 struct lustre_md *md)
1227 struct ll_sb_info *sbi = ll_s2sbi(sb);
1228 struct mdt_body *body = md->body;
1229 struct inode *inode;
1233 ino = cl_fid_build_ino(fid, sbi->ll_flags & LL_SBI_32BIT_API);
1234 inode = iget_locked(sb, ino);
1235 if (inode == NULL) {
1236 CERROR("%s: failed get simple inode "DFID": rc = -ENOENT\n",
1237 ll_get_fsname(sb, NULL, 0), PFID(fid));
1238 RETURN(ERR_PTR(-ENOENT));
1241 if (inode->i_state & I_NEW) {
1242 struct ll_inode_info *lli = ll_i2info(inode);
1243 struct lmv_stripe_md *lsm = md->lmv;
1245 inode->i_mode = (inode->i_mode & ~S_IFMT) |
1246 (body->mbo_mode & S_IFMT);
1247 LASSERTF(S_ISDIR(inode->i_mode), "Not slave inode "DFID"\n",
1250 LTIME_S(inode->i_mtime) = 0;
1251 LTIME_S(inode->i_atime) = 0;
1252 LTIME_S(inode->i_ctime) = 0;
1255 #ifdef HAVE_BACKING_DEV_INFO
1256 /* initializing backing dev info. */
1257 inode->i_mapping->backing_dev_info =
1258 &s2lsi(inode->i_sb)->lsi_bdi;
1260 inode->i_op = &ll_dir_inode_operations;
1261 inode->i_fop = &ll_dir_operations;
1262 lli->lli_fid = *fid;
1265 LASSERT(lsm != NULL);
1266 /* master object FID */
1267 lli->lli_pfid = body->mbo_fid1;
1268 CDEBUG(D_INODE, "lli %p slave "DFID" master "DFID"\n",
1269 lli, PFID(fid), PFID(&lli->lli_pfid));
1270 unlock_new_inode(inode);
1276 static int ll_init_lsm_md(struct inode *inode, struct lustre_md *md)
1279 struct lmv_stripe_md *lsm = md->lmv;
1282 LASSERT(lsm != NULL);
1283 /* XXX sigh, this lsm_root initialization should be in
1284 * LMV layer, but it needs ll_iget right now, so we
1285 * put this here right now. */
1286 for (i = 0; i < lsm->lsm_md_stripe_count; i++) {
1287 fid = &lsm->lsm_md_oinfo[i].lmo_fid;
1288 LASSERT(lsm->lsm_md_oinfo[i].lmo_root == NULL);
1289 /* Unfortunately ll_iget will call ll_update_inode,
1290 * where the initialization of slave inode is slightly
1291 * different, so it reset lsm_md to NULL to avoid
1292 * initializing lsm for slave inode. */
1293 /* For migrating inode, master stripe and master object will
1294 * be same, so we only need assign this inode */
1295 if (lsm->lsm_md_hash_type & LMV_HASH_FLAG_MIGRATION && i == 0)
1296 lsm->lsm_md_oinfo[i].lmo_root = inode;
1298 lsm->lsm_md_oinfo[i].lmo_root =
1299 ll_iget_anon_dir(inode->i_sb, fid, md);
1301 if (IS_ERR(lsm->lsm_md_oinfo[i].lmo_root)) {
1302 int rc = PTR_ERR(lsm->lsm_md_oinfo[i].lmo_root);
1304 lsm->lsm_md_oinfo[i].lmo_root = NULL;
1312 static inline int lli_lsm_md_eq(const struct lmv_stripe_md *lsm_md1,
1313 const struct lmv_stripe_md *lsm_md2)
1315 return lsm_md1->lsm_md_magic == lsm_md2->lsm_md_magic &&
1316 lsm_md1->lsm_md_stripe_count == lsm_md2->lsm_md_stripe_count &&
1317 lsm_md1->lsm_md_master_mdt_index ==
1318 lsm_md2->lsm_md_master_mdt_index &&
1319 lsm_md1->lsm_md_hash_type == lsm_md2->lsm_md_hash_type &&
1320 lsm_md1->lsm_md_layout_version ==
1321 lsm_md2->lsm_md_layout_version &&
1322 strcmp(lsm_md1->lsm_md_pool_name,
1323 lsm_md2->lsm_md_pool_name) == 0;
1326 static int ll_update_lsm_md(struct inode *inode, struct lustre_md *md)
1328 struct ll_inode_info *lli = ll_i2info(inode);
1329 struct lmv_stripe_md *lsm = md->lmv;
1333 LASSERT(S_ISDIR(inode->i_mode));
1334 CDEBUG(D_INODE, "update lsm %p of "DFID"\n", lli->lli_lsm_md,
1335 PFID(ll_inode2fid(inode)));
1337 /* no striped information from request. */
1339 if (lli->lli_lsm_md == NULL) {
1341 } else if (lli->lli_lsm_md->lsm_md_hash_type &
1342 LMV_HASH_FLAG_MIGRATION) {
1343 /* migration is done, the temporay MIGRATE layout has
1345 CDEBUG(D_INODE, DFID" finish migration.\n",
1346 PFID(ll_inode2fid(inode)));
1347 lmv_free_memmd(lli->lli_lsm_md);
1348 lli->lli_lsm_md = NULL;
1351 /* The lustre_md from req does not include stripeEA,
1352 * see ll_md_setattr */
1357 /* set the directory layout */
1358 if (lli->lli_lsm_md == NULL) {
1359 struct cl_attr *attr;
1361 rc = ll_init_lsm_md(inode, md);
1365 /* set md->lmv to NULL, so the following free lustre_md
1366 * will not free this lsm */
1368 lli->lli_lsm_md = lsm;
1370 OBD_ALLOC_PTR(attr);
1374 /* validate the lsm */
1375 rc = md_merge_attr(ll_i2mdexp(inode), lsm, attr,
1376 ll_md_blocking_ast);
1382 if (md->body->mbo_valid & OBD_MD_FLNLINK)
1383 md->body->mbo_nlink = attr->cat_nlink;
1384 if (md->body->mbo_valid & OBD_MD_FLSIZE)
1385 md->body->mbo_size = attr->cat_size;
1386 if (md->body->mbo_valid & OBD_MD_FLATIME)
1387 md->body->mbo_atime = attr->cat_atime;
1388 if (md->body->mbo_valid & OBD_MD_FLCTIME)
1389 md->body->mbo_ctime = attr->cat_ctime;
1390 if (md->body->mbo_valid & OBD_MD_FLMTIME)
1391 md->body->mbo_mtime = attr->cat_mtime;
1395 CDEBUG(D_INODE, "Set lsm %p magic %x to "DFID"\n", lsm,
1396 lsm->lsm_md_magic, PFID(ll_inode2fid(inode)));
1400 /* Compare the old and new stripe information */
1401 if (!lsm_md_eq(lli->lli_lsm_md, lsm)) {
1402 struct lmv_stripe_md *old_lsm = lli->lli_lsm_md;
1405 CERROR("%s: inode "DFID"(%p)'s lmv layout mismatch (%p)/(%p)"
1406 "magic:0x%x/0x%x stripe count: %d/%d master_mdt: %d/%d"
1407 "hash_type:0x%x/0x%x layout: 0x%x/0x%x pool:%s/%s\n",
1408 ll_get_fsname(inode->i_sb, NULL, 0), PFID(&lli->lli_fid),
1409 inode, lsm, old_lsm,
1410 lsm->lsm_md_magic, old_lsm->lsm_md_magic,
1411 lsm->lsm_md_stripe_count,
1412 old_lsm->lsm_md_stripe_count,
1413 lsm->lsm_md_master_mdt_index,
1414 old_lsm->lsm_md_master_mdt_index,
1415 lsm->lsm_md_hash_type, old_lsm->lsm_md_hash_type,
1416 lsm->lsm_md_layout_version,
1417 old_lsm->lsm_md_layout_version,
1418 lsm->lsm_md_pool_name,
1419 old_lsm->lsm_md_pool_name);
1421 for (idx = 0; idx < old_lsm->lsm_md_stripe_count; idx++) {
1422 CERROR("%s: sub FIDs in old lsm idx %d, old: "DFID"\n",
1423 ll_get_fsname(inode->i_sb, NULL, 0), idx,
1424 PFID(&old_lsm->lsm_md_oinfo[idx].lmo_fid));
1427 for (idx = 0; idx < lsm->lsm_md_stripe_count; idx++) {
1428 CERROR("%s: sub FIDs in new lsm idx %d, new: "DFID"\n",
1429 ll_get_fsname(inode->i_sb, NULL, 0), idx,
1430 PFID(&lsm->lsm_md_oinfo[idx].lmo_fid));
1439 void ll_clear_inode(struct inode *inode)
1441 struct ll_inode_info *lli = ll_i2info(inode);
1442 struct ll_sb_info *sbi = ll_i2sbi(inode);
1445 CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p)\n",
1446 PFID(ll_inode2fid(inode)), inode);
1448 if (S_ISDIR(inode->i_mode)) {
1449 /* these should have been cleared in ll_file_release */
1450 LASSERT(lli->lli_opendir_key == NULL);
1451 LASSERT(lli->lli_sai == NULL);
1452 LASSERT(lli->lli_opendir_pid == 0);
1455 md_null_inode(sbi->ll_md_exp, ll_inode2fid(inode));
1457 LASSERT(!lli->lli_open_fd_write_count);
1458 LASSERT(!lli->lli_open_fd_read_count);
1459 LASSERT(!lli->lli_open_fd_exec_count);
1461 if (lli->lli_mds_write_och)
1462 ll_md_real_close(inode, FMODE_WRITE);
1463 if (lli->lli_mds_exec_och)
1464 ll_md_real_close(inode, FMODE_EXEC);
1465 if (lli->lli_mds_read_och)
1466 ll_md_real_close(inode, FMODE_READ);
1468 if (S_ISLNK(inode->i_mode) && lli->lli_symlink_name) {
1469 OBD_FREE(lli->lli_symlink_name,
1470 strlen(lli->lli_symlink_name) + 1);
1471 lli->lli_symlink_name = NULL;
1474 ll_xattr_cache_destroy(inode);
1476 #ifdef CONFIG_FS_POSIX_ACL
1477 forget_all_cached_acls(inode);
1478 if (lli->lli_posix_acl) {
1479 LASSERT(atomic_read(&lli->lli_posix_acl->a_refcount) == 1);
1480 posix_acl_release(lli->lli_posix_acl);
1481 lli->lli_posix_acl = NULL;
1484 lli->lli_inode_magic = LLI_INODE_DEAD;
1486 if (S_ISDIR(inode->i_mode))
1487 ll_dir_clear_lsm_md(inode);
1488 else if (S_ISREG(inode->i_mode) && !is_bad_inode(inode))
1489 LASSERT(list_empty(&lli->lli_agl_list));
1492 * XXX This has to be done before lsm is freed below, because
1493 * cl_object still uses inode lsm.
1495 cl_inode_fini(inode);
1500 static int ll_md_setattr(struct dentry *dentry, struct md_op_data *op_data)
1502 struct lustre_md md;
1503 struct inode *inode = dentry->d_inode;
1504 struct ll_sb_info *sbi = ll_i2sbi(inode);
1505 struct ptlrpc_request *request = NULL;
1509 op_data = ll_prep_md_op_data(op_data, inode, NULL, NULL, 0, 0,
1510 LUSTRE_OPC_ANY, NULL);
1511 if (IS_ERR(op_data))
1512 RETURN(PTR_ERR(op_data));
1514 rc = md_setattr(sbi->ll_md_exp, op_data, NULL, 0, &request);
1516 ptlrpc_req_finished(request);
1517 if (rc == -ENOENT) {
1519 /* Unlinked special device node? Or just a race?
1520 * Pretend we done everything. */
1521 if (!S_ISREG(inode->i_mode) &&
1522 !S_ISDIR(inode->i_mode)) {
1523 ia_valid = op_data->op_attr.ia_valid;
1524 op_data->op_attr.ia_valid &= ~TIMES_SET_FLAGS;
1525 rc = simple_setattr(dentry, &op_data->op_attr);
1526 op_data->op_attr.ia_valid = ia_valid;
1528 } else if (rc != -EPERM && rc != -EACCES && rc != -ETXTBSY) {
1529 CERROR("md_setattr fails: rc = %d\n", rc);
1534 rc = md_get_lustre_md(sbi->ll_md_exp, request, sbi->ll_dt_exp,
1535 sbi->ll_md_exp, &md);
1537 ptlrpc_req_finished(request);
1541 ia_valid = op_data->op_attr.ia_valid;
1542 /* inode size will be in ll_setattr_ost, can't do it now since dirty
1543 * cache is not cleared yet. */
1544 op_data->op_attr.ia_valid &= ~(TIMES_SET_FLAGS | ATTR_SIZE);
1545 if (S_ISREG(inode->i_mode))
1547 rc = simple_setattr(dentry, &op_data->op_attr);
1548 if (S_ISREG(inode->i_mode))
1549 inode_unlock(inode);
1550 op_data->op_attr.ia_valid = ia_valid;
1552 rc = ll_update_inode(inode, &md);
1553 ptlrpc_req_finished(request);
1558 /* If this inode has objects allocated to it (lsm != NULL), then the OST
1559 * object(s) determine the file size and mtime. Otherwise, the MDS will
1560 * keep these values until such a time that objects are allocated for it.
1561 * We do the MDS operations first, as it is checking permissions for us.
1562 * We don't to the MDS RPC if there is nothing that we want to store there,
1563 * otherwise there is no harm in updating mtime/atime on the MDS if we are
1564 * going to do an RPC anyways.
1566 * If we are doing a truncate, we will send the mtime and ctime updates
1567 * to the OST with the punch RPC, otherwise we do an explicit setattr RPC.
1568 * I don't believe it is possible to get e.g. ATTR_MTIME_SET and ATTR_SIZE
1571 * In case of HSMimport, we only set attr on MDS.
1573 int ll_setattr_raw(struct dentry *dentry, struct iattr *attr, bool hsm_import)
1575 struct inode *inode = dentry->d_inode;
1576 struct ll_inode_info *lli = ll_i2info(inode);
1577 struct md_op_data *op_data = NULL;
1581 CDEBUG(D_VFSTRACE, "%s: setattr inode "DFID"(%p) from %llu to %llu, "
1582 "valid %x, hsm_import %d\n",
1583 ll_get_fsname(inode->i_sb, NULL, 0), PFID(&lli->lli_fid),
1584 inode, i_size_read(inode), attr->ia_size, attr->ia_valid,
1587 if (attr->ia_valid & ATTR_SIZE) {
1588 /* Check new size against VFS/VM file size limit and rlimit */
1589 rc = inode_newsize_ok(inode, attr->ia_size);
1593 /* The maximum Lustre file size is variable, based on the
1594 * OST maximum object size and number of stripes. This
1595 * needs another check in addition to the VFS check above. */
1596 if (attr->ia_size > ll_file_maxbytes(inode)) {
1597 CDEBUG(D_INODE,"file "DFID" too large %llu > %llu\n",
1598 PFID(&lli->lli_fid), attr->ia_size,
1599 ll_file_maxbytes(inode));
1603 attr->ia_valid |= ATTR_MTIME | ATTR_CTIME;
1606 /* POSIX: check before ATTR_*TIME_SET set (from inode_change_ok) */
1607 if (attr->ia_valid & TIMES_SET_FLAGS) {
1608 if ((!uid_eq(current_fsuid(), inode->i_uid)) &&
1609 !cfs_capable(CFS_CAP_FOWNER))
1613 /* We mark all of the fields "set" so MDS/OST does not re-set them */
1614 if (!(attr->ia_valid & ATTR_CTIME_SET) &&
1615 (attr->ia_valid & ATTR_CTIME)) {
1616 attr->ia_ctime = current_time(inode);
1617 attr->ia_valid |= ATTR_CTIME_SET;
1619 if (!(attr->ia_valid & ATTR_ATIME_SET) &&
1620 (attr->ia_valid & ATTR_ATIME)) {
1621 attr->ia_atime = current_time(inode);
1622 attr->ia_valid |= ATTR_ATIME_SET;
1624 if (!(attr->ia_valid & ATTR_MTIME_SET) &&
1625 (attr->ia_valid & ATTR_MTIME)) {
1626 attr->ia_mtime = current_time(inode);
1627 attr->ia_valid |= ATTR_MTIME_SET;
1630 if (attr->ia_valid & (ATTR_MTIME | ATTR_CTIME))
1631 CDEBUG(D_INODE, "setting mtime %lu, ctime %lu, now = %llu\n",
1632 LTIME_S(attr->ia_mtime), LTIME_S(attr->ia_ctime),
1633 (s64)ktime_get_real_seconds());
1635 if (S_ISREG(inode->i_mode)) {
1636 if (attr->ia_valid & ATTR_SIZE)
1637 inode_dio_write_done(inode);
1638 inode_unlock(inode);
1641 /* We always do an MDS RPC, even if we're only changing the size;
1642 * only the MDS knows whether truncate() should fail with -ETXTBUSY */
1644 OBD_ALLOC_PTR(op_data);
1645 if (op_data == NULL)
1646 GOTO(out, rc = -ENOMEM);
1648 if (!hsm_import && attr->ia_valid & ATTR_SIZE) {
1649 /* If we are changing file size, file content is
1650 * modified, flag it. */
1651 attr->ia_valid |= MDS_OPEN_OWNEROVERRIDE;
1652 op_data->op_bias |= MDS_DATA_MODIFIED;
1653 ll_file_clear_flag(lli, LLIF_DATA_MODIFIED);
1656 op_data->op_attr = *attr;
1658 rc = ll_md_setattr(dentry, op_data);
1662 if (!S_ISREG(inode->i_mode) || hsm_import)
1665 if (attr->ia_valid & (ATTR_SIZE |
1666 ATTR_ATIME | ATTR_ATIME_SET |
1667 ATTR_MTIME | ATTR_MTIME_SET |
1668 ATTR_CTIME | ATTR_CTIME_SET)) {
1669 /* For truncate and utimes sending attributes to OSTs, setting
1670 * mtime/atime to the past will be performed under PW [0:EOF]
1671 * extent lock (new_size:EOF for truncate). It may seem
1672 * excessive to send mtime/atime updates to OSTs when not
1673 * setting times to past, but it is necessary due to possible
1674 * time de-synchronization between MDT inode and OST objects */
1675 rc = cl_setattr_ost(lli->lli_clob, attr, 0);
1678 /* If the file was restored, it needs to set dirty flag.
1680 * We've already sent MDS_DATA_MODIFIED flag in
1681 * ll_md_setattr() for truncate. However, the MDT refuses to
1682 * set the HS_DIRTY flag on released files, so we have to set
1683 * it again if the file has been restored. Please check how
1684 * LLIF_DATA_MODIFIED is set in vvp_io_setattr_fini().
1686 * Please notice that if the file is not released, the previous
1687 * MDS_DATA_MODIFIED has taken effect and usually
1688 * LLIF_DATA_MODIFIED is not set(see vvp_io_setattr_fini()).
1689 * This way we can save an RPC for common open + trunc
1691 if (ll_file_test_and_clear_flag(lli, LLIF_DATA_MODIFIED)) {
1692 struct hsm_state_set hss = {
1693 .hss_valid = HSS_SETMASK,
1694 .hss_setmask = HS_DIRTY,
1698 rc2 = ll_hsm_state_set(inode, &hss);
1699 /* truncate and write can happen at the same time, so that
1700 * the file can be set modified even though the file is not
1701 * restored from released state, and ll_hsm_state_set() is
1702 * not applicable for the file, and rc2 < 0 is normal in this
1705 CDEBUG(D_INFO, DFID "HSM set dirty failed: rc2 = %d\n",
1706 PFID(ll_inode2fid(inode)), rc2);
1711 if (op_data != NULL)
1712 ll_finish_md_op_data(op_data);
1714 if (S_ISREG(inode->i_mode)) {
1716 if ((attr->ia_valid & ATTR_SIZE) && !hsm_import)
1717 inode_dio_wait(inode);
1718 /* Once we've got the i_mutex, it's safe to set the S_NOSEC
1719 * flag. ll_update_inode (called from ll_md_setattr), clears
1720 * inode flags, so there is a gap where S_NOSEC is not set.
1721 * This can cause a writer to take the i_mutex unnecessarily,
1722 * but this is safe to do and should be rare. */
1723 inode_has_no_xattr(inode);
1726 ll_stats_ops_tally(ll_i2sbi(inode), (attr->ia_valid & ATTR_SIZE) ?
1727 LPROC_LL_TRUNC : LPROC_LL_SETATTR, 1);
1732 int ll_setattr(struct dentry *de, struct iattr *attr)
1734 int mode = de->d_inode->i_mode;
1736 if ((attr->ia_valid & (ATTR_CTIME|ATTR_SIZE|ATTR_MODE)) ==
1737 (ATTR_CTIME|ATTR_SIZE|ATTR_MODE))
1738 attr->ia_valid |= MDS_OPEN_OWNEROVERRIDE;
1740 if (((attr->ia_valid & (ATTR_MODE|ATTR_FORCE|ATTR_SIZE)) ==
1741 (ATTR_SIZE|ATTR_MODE)) &&
1742 (((mode & S_ISUID) && !(attr->ia_mode & S_ISUID)) ||
1743 (((mode & (S_ISGID|S_IXGRP)) == (S_ISGID|S_IXGRP)) &&
1744 !(attr->ia_mode & S_ISGID))))
1745 attr->ia_valid |= ATTR_FORCE;
1747 if ((attr->ia_valid & ATTR_MODE) &&
1749 !(attr->ia_mode & S_ISUID) &&
1750 !(attr->ia_valid & ATTR_KILL_SUID))
1751 attr->ia_valid |= ATTR_KILL_SUID;
1753 if ((attr->ia_valid & ATTR_MODE) &&
1754 ((mode & (S_ISGID|S_IXGRP)) == (S_ISGID|S_IXGRP)) &&
1755 !(attr->ia_mode & S_ISGID) &&
1756 !(attr->ia_valid & ATTR_KILL_SGID))
1757 attr->ia_valid |= ATTR_KILL_SGID;
1759 /* avoid polluted from ATTR_TIMES_SET,
1760 * projid is not expected to be set here */
1761 attr->ia_valid &= ~MDS_ATTR_PROJID;
1763 return ll_setattr_raw(de, attr, false);
1766 int ll_statfs_internal(struct super_block *sb, struct obd_statfs *osfs,
1767 __u64 max_age, __u32 flags)
1769 struct ll_sb_info *sbi = ll_s2sbi(sb);
1770 struct obd_statfs obd_osfs;
1774 rc = obd_statfs(NULL, sbi->ll_md_exp, osfs, max_age, flags);
1776 CERROR("md_statfs fails: rc = %d\n", rc);
1780 osfs->os_type = sb->s_magic;
1782 CDEBUG(D_SUPER, "MDC blocks %llu/%llu objects %llu/%llu\n",
1783 osfs->os_bavail, osfs->os_blocks, osfs->os_ffree,osfs->os_files);
1785 if (sbi->ll_flags & LL_SBI_LAZYSTATFS)
1786 flags |= OBD_STATFS_NODELAY;
1788 rc = obd_statfs_rqset(sbi->ll_dt_exp, &obd_osfs, max_age, flags);
1790 CERROR("obd_statfs fails: rc = %d\n", rc);
1794 CDEBUG(D_SUPER, "OSC blocks %llu/%llu objects %llu/%llu\n",
1795 obd_osfs.os_bavail, obd_osfs.os_blocks, obd_osfs.os_ffree,
1798 osfs->os_bsize = obd_osfs.os_bsize;
1799 osfs->os_blocks = obd_osfs.os_blocks;
1800 osfs->os_bfree = obd_osfs.os_bfree;
1801 osfs->os_bavail = obd_osfs.os_bavail;
1803 /* If we don't have as many objects free on the OST as inodes
1804 * on the MDS, we reduce the total number of inodes to
1805 * compensate, so that the "inodes in use" number is correct.
1807 if (obd_osfs.os_ffree < osfs->os_ffree) {
1808 osfs->os_files = (osfs->os_files - osfs->os_ffree) +
1810 osfs->os_ffree = obd_osfs.os_ffree;
1815 int ll_statfs(struct dentry *de, struct kstatfs *sfs)
1817 struct super_block *sb = de->d_sb;
1818 struct obd_statfs osfs;
1819 __u64 fsid = huge_encode_dev(sb->s_dev);
1822 CDEBUG(D_VFSTRACE, "VFS Op: at %llu jiffies\n", get_jiffies_64());
1823 ll_stats_ops_tally(ll_s2sbi(sb), LPROC_LL_STAFS, 1);
1825 /* Some amount of caching on the client is allowed */
1826 rc = ll_statfs_internal(sb, &osfs,
1827 cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS),
1832 statfs_unpack(sfs, &osfs);
1834 /* We need to downshift for all 32-bit kernels, because we can't
1835 * tell if the kernel is being called via sys_statfs64() or not.
1836 * Stop before overflowing f_bsize - in which case it is better
1837 * to just risk EOVERFLOW if caller is using old sys_statfs(). */
1838 if (sizeof(long) < 8) {
1839 while (osfs.os_blocks > ~0UL && sfs->f_bsize < 0x40000000) {
1842 osfs.os_blocks >>= 1;
1843 osfs.os_bfree >>= 1;
1844 osfs.os_bavail >>= 1;
1848 sfs->f_blocks = osfs.os_blocks;
1849 sfs->f_bfree = osfs.os_bfree;
1850 sfs->f_bavail = osfs.os_bavail;
1851 sfs->f_fsid.val[0] = (__u32)fsid;
1852 sfs->f_fsid.val[1] = (__u32)(fsid >> 32);
1856 void ll_inode_size_lock(struct inode *inode)
1858 struct ll_inode_info *lli;
1860 LASSERT(!S_ISDIR(inode->i_mode));
1862 lli = ll_i2info(inode);
1863 mutex_lock(&lli->lli_size_mutex);
1866 void ll_inode_size_unlock(struct inode *inode)
1868 struct ll_inode_info *lli;
1870 lli = ll_i2info(inode);
1871 mutex_unlock(&lli->lli_size_mutex);
1874 int ll_update_inode(struct inode *inode, struct lustre_md *md)
1876 struct ll_inode_info *lli = ll_i2info(inode);
1877 struct mdt_body *body = md->body;
1878 struct ll_sb_info *sbi = ll_i2sbi(inode);
1881 if (body->mbo_valid & OBD_MD_FLEASIZE) {
1882 rc = cl_file_inode_init(inode, md);
1887 if (S_ISDIR(inode->i_mode)) {
1888 rc = ll_update_lsm_md(inode, md);
1893 #ifdef CONFIG_FS_POSIX_ACL
1894 if (body->mbo_valid & OBD_MD_FLACL) {
1895 spin_lock(&lli->lli_lock);
1896 if (lli->lli_posix_acl)
1897 posix_acl_release(lli->lli_posix_acl);
1898 lli->lli_posix_acl = md->posix_acl;
1899 spin_unlock(&lli->lli_lock);
1902 inode->i_ino = cl_fid_build_ino(&body->mbo_fid1,
1903 sbi->ll_flags & LL_SBI_32BIT_API);
1904 inode->i_generation = cl_fid_build_gen(&body->mbo_fid1);
1906 if (body->mbo_valid & OBD_MD_FLATIME) {
1907 if (body->mbo_atime > LTIME_S(inode->i_atime))
1908 LTIME_S(inode->i_atime) = body->mbo_atime;
1909 lli->lli_atime = body->mbo_atime;
1912 if (body->mbo_valid & OBD_MD_FLMTIME) {
1913 if (body->mbo_mtime > LTIME_S(inode->i_mtime)) {
1914 CDEBUG(D_INODE, "setting ino %lu mtime from %lu "
1915 "to %llu\n", inode->i_ino,
1916 LTIME_S(inode->i_mtime), body->mbo_mtime);
1917 LTIME_S(inode->i_mtime) = body->mbo_mtime;
1919 lli->lli_mtime = body->mbo_mtime;
1922 if (body->mbo_valid & OBD_MD_FLCTIME) {
1923 if (body->mbo_ctime > LTIME_S(inode->i_ctime))
1924 LTIME_S(inode->i_ctime) = body->mbo_ctime;
1925 lli->lli_ctime = body->mbo_ctime;
1928 /* Clear i_flags to remove S_NOSEC before permissions are updated */
1929 if (body->mbo_valid & OBD_MD_FLFLAGS)
1930 inode->i_flags = ll_ext_to_inode_flags(body->mbo_flags);
1931 if (body->mbo_valid & OBD_MD_FLMODE)
1932 inode->i_mode = (inode->i_mode & S_IFMT) |
1933 (body->mbo_mode & ~S_IFMT);
1935 if (body->mbo_valid & OBD_MD_FLTYPE)
1936 inode->i_mode = (inode->i_mode & ~S_IFMT) |
1937 (body->mbo_mode & S_IFMT);
1939 LASSERT(inode->i_mode != 0);
1940 if (S_ISREG(inode->i_mode))
1941 inode->i_blkbits = min(PTLRPC_MAX_BRW_BITS + 1,
1942 LL_MAX_BLKSIZE_BITS);
1944 inode->i_blkbits = inode->i_sb->s_blocksize_bits;
1946 if (body->mbo_valid & OBD_MD_FLUID)
1947 inode->i_uid = make_kuid(&init_user_ns, body->mbo_uid);
1948 if (body->mbo_valid & OBD_MD_FLGID)
1949 inode->i_gid = make_kgid(&init_user_ns, body->mbo_gid);
1950 if (body->mbo_valid & OBD_MD_FLPROJID)
1951 lli->lli_projid = body->mbo_projid;
1952 if (body->mbo_valid & OBD_MD_FLNLINK)
1953 set_nlink(inode, body->mbo_nlink);
1954 if (body->mbo_valid & OBD_MD_FLRDEV)
1955 inode->i_rdev = old_decode_dev(body->mbo_rdev);
1957 if (body->mbo_valid & OBD_MD_FLID) {
1958 /* FID shouldn't be changed! */
1959 if (fid_is_sane(&lli->lli_fid)) {
1960 LASSERTF(lu_fid_eq(&lli->lli_fid, &body->mbo_fid1),
1961 "Trying to change FID "DFID
1962 " to the "DFID", inode "DFID"(%p)\n",
1963 PFID(&lli->lli_fid), PFID(&body->mbo_fid1),
1964 PFID(ll_inode2fid(inode)), inode);
1966 lli->lli_fid = body->mbo_fid1;
1970 LASSERT(fid_seq(&lli->lli_fid) != 0);
1972 if (body->mbo_valid & OBD_MD_FLSIZE) {
1973 i_size_write(inode, body->mbo_size);
1975 CDEBUG(D_VFSTRACE, "inode="DFID", updating i_size %llu\n",
1976 PFID(ll_inode2fid(inode)),
1977 (unsigned long long)body->mbo_size);
1979 if (body->mbo_valid & OBD_MD_FLBLOCKS)
1980 inode->i_blocks = body->mbo_blocks;
1983 if (body->mbo_valid & OBD_MD_TSTATE) {
1984 /* Set LLIF_FILE_RESTORING if restore ongoing and
1985 * clear it when done to ensure to start again
1986 * glimpsing updated attrs
1988 if (body->mbo_t_state & MS_RESTORE)
1989 ll_file_set_flag(lli, LLIF_FILE_RESTORING);
1991 ll_file_clear_flag(lli, LLIF_FILE_RESTORING);
1997 int ll_read_inode2(struct inode *inode, void *opaque)
1999 struct lustre_md *md = opaque;
2000 struct ll_inode_info *lli = ll_i2info(inode);
2004 CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p)\n",
2005 PFID(&lli->lli_fid), inode);
2007 /* Core attributes from the MDS first. This is a new inode, and
2008 * the VFS doesn't zero times in the core inode so we have to do
2009 * it ourselves. They will be overwritten by either MDS or OST
2010 * attributes - we just need to make sure they aren't newer. */
2011 LTIME_S(inode->i_mtime) = 0;
2012 LTIME_S(inode->i_atime) = 0;
2013 LTIME_S(inode->i_ctime) = 0;
2015 rc = ll_update_inode(inode, md);
2019 /* OIDEBUG(inode); */
2021 #ifdef HAVE_BACKING_DEV_INFO
2022 /* initializing backing dev info. */
2023 inode->i_mapping->backing_dev_info = &s2lsi(inode->i_sb)->lsi_bdi;
2025 if (S_ISREG(inode->i_mode)) {
2026 struct ll_sb_info *sbi = ll_i2sbi(inode);
2027 inode->i_op = &ll_file_inode_operations;
2028 inode->i_fop = sbi->ll_fop;
2029 inode->i_mapping->a_ops = (struct address_space_operations *)&ll_aops;
2031 } else if (S_ISDIR(inode->i_mode)) {
2032 inode->i_op = &ll_dir_inode_operations;
2033 inode->i_fop = &ll_dir_operations;
2035 } else if (S_ISLNK(inode->i_mode)) {
2036 inode->i_op = &ll_fast_symlink_inode_operations;
2039 inode->i_op = &ll_special_inode_operations;
2041 init_special_inode(inode, inode->i_mode,
2050 void ll_delete_inode(struct inode *inode)
2052 struct ll_inode_info *lli = ll_i2info(inode);
2055 if (S_ISREG(inode->i_mode) && lli->lli_clob != NULL)
2056 /* It is last chance to write out dirty pages,
2057 * otherwise we may lose data while umount */
2058 cl_sync_file_range(inode, 0, OBD_OBJECT_EOF, CL_FSYNC_LOCAL, 1);
2060 truncate_inode_pages_final(&inode->i_data);
2062 LASSERTF(inode->i_data.nrpages == 0, "inode="DFID"(%p) nrpages=%lu, "
2063 "see https://jira.hpdd.intel.com/browse/LU-118\n",
2064 PFID(ll_inode2fid(inode)), inode, inode->i_data.nrpages);
2066 #ifdef HAVE_SBOPS_EVICT_INODE
2067 ll_clear_inode(inode);
2074 int ll_iocontrol(struct inode *inode, struct file *file,
2075 unsigned int cmd, unsigned long arg)
2077 struct ll_sb_info *sbi = ll_i2sbi(inode);
2078 struct ptlrpc_request *req = NULL;
2083 case FSFILT_IOC_GETFLAGS: {
2084 struct mdt_body *body;
2085 struct md_op_data *op_data;
2087 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL,
2088 0, 0, LUSTRE_OPC_ANY,
2090 if (IS_ERR(op_data))
2091 RETURN(PTR_ERR(op_data));
2093 op_data->op_valid = OBD_MD_FLFLAGS;
2094 rc = md_getattr(sbi->ll_md_exp, op_data, &req);
2095 ll_finish_md_op_data(op_data);
2097 CERROR("%s: failure inode "DFID": rc = %d\n",
2098 sbi->ll_md_exp->exp_obd->obd_name,
2099 PFID(ll_inode2fid(inode)), rc);
2103 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
2105 flags = body->mbo_flags;
2107 ptlrpc_req_finished(req);
2109 RETURN(put_user(flags, (int __user *)arg));
2111 case FSFILT_IOC_SETFLAGS: {
2113 struct md_op_data *op_data;
2114 struct cl_object *obj;
2116 if (get_user(flags, (int __user *)arg))
2119 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
2120 LUSTRE_OPC_ANY, NULL);
2121 if (IS_ERR(op_data))
2122 RETURN(PTR_ERR(op_data));
2124 op_data->op_attr_flags = flags;
2125 op_data->op_attr.ia_valid |= ATTR_ATTR_FLAG;
2126 rc = md_setattr(sbi->ll_md_exp, op_data, NULL, 0, &req);
2127 ll_finish_md_op_data(op_data);
2128 ptlrpc_req_finished(req);
2132 inode->i_flags = ll_ext_to_inode_flags(flags);
2134 obj = ll_i2info(inode)->lli_clob;
2138 OBD_ALLOC_PTR(attr);
2142 attr->ia_valid = ATTR_ATTR_FLAG;
2143 rc = cl_setattr_ost(obj, attr, flags);
2155 int ll_flush_ctx(struct inode *inode)
2157 struct ll_sb_info *sbi = ll_i2sbi(inode);
2159 CDEBUG(D_SEC, "flush context for user %d\n",
2160 from_kuid(&init_user_ns, current_uid()));
2162 obd_set_info_async(NULL, sbi->ll_md_exp,
2163 sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
2165 obd_set_info_async(NULL, sbi->ll_dt_exp,
2166 sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
2171 /* umount -f client means force down, don't save state */
2172 void ll_umount_begin(struct super_block *sb)
2174 struct ll_sb_info *sbi = ll_s2sbi(sb);
2175 struct obd_device *obd;
2176 struct obd_ioctl_data *ioc_data;
2177 struct l_wait_info lwi;
2178 wait_queue_head_t waitq;
2181 CDEBUG(D_VFSTRACE, "VFS Op: superblock %p count %d active %d\n", sb,
2182 sb->s_count, atomic_read(&sb->s_active));
2184 obd = class_exp2obd(sbi->ll_md_exp);
2186 CERROR("Invalid MDC connection handle %#llx\n",
2187 sbi->ll_md_exp->exp_handle.h_cookie);
2193 obd = class_exp2obd(sbi->ll_dt_exp);
2195 CERROR("Invalid LOV connection handle %#llx\n",
2196 sbi->ll_dt_exp->exp_handle.h_cookie);
2202 OBD_ALLOC_PTR(ioc_data);
2204 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_md_exp,
2205 sizeof *ioc_data, ioc_data, NULL);
2207 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_dt_exp,
2208 sizeof *ioc_data, ioc_data, NULL);
2210 OBD_FREE_PTR(ioc_data);
2213 /* Really, we'd like to wait until there are no requests outstanding,
2214 * and then continue. For now, we just periodically checking for vfs
2215 * to decrement mnt_cnt and hope to finish it within 10sec.
2217 init_waitqueue_head(&waitq);
2218 lwi = LWI_TIMEOUT_INTERVAL(cfs_time_seconds(10),
2219 cfs_time_seconds(1), NULL, NULL);
2220 l_wait_event(waitq, may_umount(sbi->ll_mnt.mnt), &lwi);
2225 int ll_remount_fs(struct super_block *sb, int *flags, char *data)
2227 struct ll_sb_info *sbi = ll_s2sbi(sb);
2228 char *profilenm = get_profile_name(sb);
2232 if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY)) {
2233 read_only = *flags & MS_RDONLY;
2234 err = obd_set_info_async(NULL, sbi->ll_md_exp,
2235 sizeof(KEY_READ_ONLY),
2236 KEY_READ_ONLY, sizeof(read_only),
2239 LCONSOLE_WARN("Failed to remount %s %s (%d)\n",
2240 profilenm, read_only ?
2241 "read-only" : "read-write", err);
2246 sb->s_flags |= MS_RDONLY;
2248 sb->s_flags &= ~MS_RDONLY;
2250 if (sbi->ll_flags & LL_SBI_VERBOSE)
2251 LCONSOLE_WARN("Remounted %s %s\n", profilenm,
2252 read_only ? "read-only" : "read-write");
2258 * Cleanup the open handle that is cached on MDT-side.
2260 * For open case, the client side open handling thread may hit error
2261 * after the MDT grant the open. Under such case, the client should
2262 * send close RPC to the MDT as cleanup; otherwise, the open handle
2263 * on the MDT will be leaked there until the client umount or evicted.
2265 * In further, if someone unlinked the file, because the open handle
2266 * holds the reference on such file/object, then it will block the
2267 * subsequent threads that want to locate such object via FID.
2269 * \param[in] sb super block for this file-system
2270 * \param[in] open_req pointer to the original open request
2272 void ll_open_cleanup(struct super_block *sb, struct ptlrpc_request *open_req)
2274 struct mdt_body *body;
2275 struct md_op_data *op_data;
2276 struct ptlrpc_request *close_req = NULL;
2277 struct obd_export *exp = ll_s2sbi(sb)->ll_md_exp;
2280 body = req_capsule_server_get(&open_req->rq_pill, &RMF_MDT_BODY);
2281 OBD_ALLOC_PTR(op_data);
2282 if (op_data == NULL) {
2283 CWARN("%s: cannot allocate op_data to release open handle for "
2285 ll_get_fsname(sb, NULL, 0), PFID(&body->mbo_fid1));
2290 op_data->op_fid1 = body->mbo_fid1;
2291 op_data->op_handle = body->mbo_handle;
2292 op_data->op_mod_time = ktime_get_real_seconds();
2293 md_close(exp, op_data, NULL, &close_req);
2294 ptlrpc_req_finished(close_req);
2295 ll_finish_md_op_data(op_data);
2300 int ll_prep_inode(struct inode **inode, struct ptlrpc_request *req,
2301 struct super_block *sb, struct lookup_intent *it)
2303 struct ll_sb_info *sbi = NULL;
2304 struct lustre_md md = { NULL };
2308 LASSERT(*inode || sb);
2309 sbi = sb ? ll_s2sbi(sb) : ll_i2sbi(*inode);
2310 rc = md_get_lustre_md(sbi->ll_md_exp, req, sbi->ll_dt_exp,
2311 sbi->ll_md_exp, &md);
2316 rc = ll_update_inode(*inode, &md);
2320 LASSERT(sb != NULL);
2323 * At this point server returns to client's same fid as client
2324 * generated for creating. So using ->fid1 is okay here.
2326 if (!fid_is_sane(&md.body->mbo_fid1)) {
2327 CERROR("%s: Fid is insane "DFID"\n",
2328 ll_get_fsname(sb, NULL, 0),
2329 PFID(&md.body->mbo_fid1));
2330 GOTO(out, rc = -EINVAL);
2333 *inode = ll_iget(sb, cl_fid_build_ino(&md.body->mbo_fid1,
2334 sbi->ll_flags & LL_SBI_32BIT_API),
2336 if (IS_ERR(*inode)) {
2337 #ifdef CONFIG_FS_POSIX_ACL
2339 posix_acl_release(md.posix_acl);
2340 md.posix_acl = NULL;
2343 rc = IS_ERR(*inode) ? PTR_ERR(*inode) : -ENOMEM;
2345 CERROR("new_inode -fatal: rc %d\n", rc);
2350 /* Handling piggyback layout lock.
2351 * Layout lock can be piggybacked by getattr and open request.
2352 * The lsm can be applied to inode only if it comes with a layout lock
2353 * otherwise correct layout may be overwritten, for example:
2354 * 1. proc1: mdt returns a lsm but not granting layout
2355 * 2. layout was changed by another client
2356 * 3. proc2: refresh layout and layout lock granted
2357 * 4. proc1: to apply a stale layout */
2358 if (it != NULL && it->it_lock_mode != 0) {
2359 struct lustre_handle lockh;
2360 struct ldlm_lock *lock;
2362 lockh.cookie = it->it_lock_handle;
2363 lock = ldlm_handle2lock(&lockh);
2364 LASSERT(lock != NULL);
2365 if (ldlm_has_layout(lock)) {
2366 struct cl_object_conf conf;
2368 memset(&conf, 0, sizeof(conf));
2369 conf.coc_opc = OBJECT_CONF_SET;
2370 conf.coc_inode = *inode;
2371 conf.coc_lock = lock;
2372 conf.u.coc_layout = md.layout;
2373 (void)ll_layout_conf(*inode, &conf);
2375 LDLM_LOCK_PUT(lock);
2381 md_free_lustre_md(sbi->ll_md_exp, &md);
2384 if (rc != 0 && it != NULL && it->it_op & IT_OPEN)
2385 ll_open_cleanup(sb != NULL ? sb : (*inode)->i_sb, req);
2390 int ll_obd_statfs(struct inode *inode, void __user *arg)
2392 struct ll_sb_info *sbi = NULL;
2393 struct obd_export *exp;
2395 struct obd_ioctl_data *data = NULL;
2399 if (!inode || !(sbi = ll_i2sbi(inode)))
2400 GOTO(out_statfs, rc = -EINVAL);
2402 rc = obd_ioctl_getdata(&buf, &len, arg);
2404 GOTO(out_statfs, rc);
2407 if (!data->ioc_inlbuf1 || !data->ioc_inlbuf2 ||
2408 !data->ioc_pbuf1 || !data->ioc_pbuf2)
2409 GOTO(out_statfs, rc = -EINVAL);
2411 if (data->ioc_inllen1 != sizeof(__u32) ||
2412 data->ioc_inllen2 != sizeof(__u32) ||
2413 data->ioc_plen1 != sizeof(struct obd_statfs) ||
2414 data->ioc_plen2 != sizeof(struct obd_uuid))
2415 GOTO(out_statfs, rc = -EINVAL);
2417 memcpy(&type, data->ioc_inlbuf1, sizeof(__u32));
2418 if (type & LL_STATFS_LMV)
2419 exp = sbi->ll_md_exp;
2420 else if (type & LL_STATFS_LOV)
2421 exp = sbi->ll_dt_exp;
2423 GOTO(out_statfs, rc = -ENODEV);
2425 rc = obd_iocontrol(IOC_OBD_STATFS, exp, len, buf, NULL);
2427 GOTO(out_statfs, rc);
2429 OBD_FREE_LARGE(buf, len);
2433 int ll_process_config(struct lustre_cfg *lcfg)
2435 struct super_block *sb;
2440 /* The instance name contains the sb: lustre-client-aacfe000 */
2441 ptr = strrchr(lustre_cfg_string(lcfg, 0), '-');
2442 if (!ptr || !*(++ptr))
2444 if (sscanf(ptr, "%lx", &x) != 1)
2446 sb = (struct super_block *)x;
2447 /* This better be a real Lustre superblock! */
2448 LASSERT(s2lsi(sb)->lsi_lmd->lmd_magic == LMD_MAGIC);
2450 /* Note we have not called client_common_fill_super yet, so
2451 proc fns must be able to handle that! */
2452 rc = class_process_proc_param(PARAM_LLITE, lprocfs_llite_obd_vars,
2459 /* this function prepares md_op_data hint for passing it down to MD stack. */
2460 struct md_op_data *ll_prep_md_op_data(struct md_op_data *op_data,
2461 struct inode *i1, struct inode *i2,
2462 const char *name, size_t namelen,
2463 __u32 mode, __u32 opc, void *data)
2465 LASSERT(i1 != NULL);
2468 /* Do not reuse namelen for something else. */
2470 return ERR_PTR(-EINVAL);
2472 if (namelen > ll_i2sbi(i1)->ll_namelen)
2473 return ERR_PTR(-ENAMETOOLONG);
2475 if (!lu_name_is_valid_2(name, namelen))
2476 return ERR_PTR(-EINVAL);
2479 if (op_data == NULL)
2480 OBD_ALLOC_PTR(op_data);
2482 if (op_data == NULL)
2483 return ERR_PTR(-ENOMEM);
2485 ll_i2gids(op_data->op_suppgids, i1, i2);
2486 op_data->op_fid1 = *ll_inode2fid(i1);
2487 op_data->op_default_stripe_offset = -1;
2488 if (S_ISDIR(i1->i_mode)) {
2489 op_data->op_mea1 = ll_i2info(i1)->lli_lsm_md;
2490 if (opc == LUSTRE_OPC_MKDIR)
2491 op_data->op_default_stripe_offset =
2492 ll_i2info(i1)->lli_def_stripe_offset;
2496 op_data->op_fid2 = *ll_inode2fid(i2);
2497 if (S_ISDIR(i2->i_mode))
2498 op_data->op_mea2 = ll_i2info(i2)->lli_lsm_md;
2500 fid_zero(&op_data->op_fid2);
2503 if (ll_i2sbi(i1)->ll_flags & LL_SBI_64BIT_HASH)
2504 op_data->op_cli_flags |= CLI_HASH64;
2506 if (ll_need_32bit_api(ll_i2sbi(i1)))
2507 op_data->op_cli_flags |= CLI_API32;
2509 op_data->op_name = name;
2510 op_data->op_namelen = namelen;
2511 op_data->op_mode = mode;
2512 op_data->op_mod_time = cfs_time_current_sec();
2513 op_data->op_fsuid = from_kuid(&init_user_ns, current_fsuid());
2514 op_data->op_fsgid = from_kgid(&init_user_ns, current_fsgid());
2515 op_data->op_cap = cfs_curproc_cap_pack();
2516 if ((opc == LUSTRE_OPC_CREATE) && (name != NULL) &&
2517 filename_is_volatile(name, namelen, &op_data->op_mds)) {
2518 op_data->op_bias |= MDS_CREATE_VOLATILE;
2520 op_data->op_mds = 0;
2522 op_data->op_data = data;
2527 void ll_finish_md_op_data(struct md_op_data *op_data)
2529 security_release_secctx(op_data->op_file_secctx,
2530 op_data->op_file_secctx_size);
2531 OBD_FREE_PTR(op_data);
2534 #ifdef HAVE_SUPEROPS_USE_DENTRY
2535 int ll_show_options(struct seq_file *seq, struct dentry *dentry)
2537 int ll_show_options(struct seq_file *seq, struct vfsmount *vfs)
2540 struct ll_sb_info *sbi;
2542 #ifdef HAVE_SUPEROPS_USE_DENTRY
2543 LASSERT((seq != NULL) && (dentry != NULL));
2544 sbi = ll_s2sbi(dentry->d_sb);
2546 LASSERT((seq != NULL) && (vfs != NULL));
2547 sbi = ll_s2sbi(vfs->mnt_sb);
2550 if (sbi->ll_flags & LL_SBI_NOLCK)
2551 seq_puts(seq, ",nolock");
2553 if (sbi->ll_flags & LL_SBI_FLOCK)
2554 seq_puts(seq, ",flock");
2556 if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
2557 seq_puts(seq, ",localflock");
2559 if (sbi->ll_flags & LL_SBI_USER_XATTR)
2560 seq_puts(seq, ",user_xattr");
2562 if (sbi->ll_flags & LL_SBI_LAZYSTATFS)
2563 seq_puts(seq, ",lazystatfs");
2565 if (sbi->ll_flags & LL_SBI_USER_FID2PATH)
2566 seq_puts(seq, ",user_fid2path");
2568 if (sbi->ll_flags & LL_SBI_ALWAYS_PING)
2569 seq_puts(seq, ",always_ping");
2575 * Get obd name by cmd, and copy out to user space
2577 int ll_get_obd_name(struct inode *inode, unsigned int cmd, unsigned long arg)
2579 struct ll_sb_info *sbi = ll_i2sbi(inode);
2580 struct obd_device *obd;
2583 if (cmd == OBD_IOC_GETDTNAME)
2584 obd = class_exp2obd(sbi->ll_dt_exp);
2585 else if (cmd == OBD_IOC_GETMDNAME)
2586 obd = class_exp2obd(sbi->ll_md_exp);
2593 if (copy_to_user((void __user *)arg, obd->obd_name,
2594 strlen(obd->obd_name) + 1))
2601 * Get lustre file system name by \a sbi. If \a buf is provided(non-NULL), the
2602 * fsname will be returned in this buffer; otherwise, a static buffer will be
2603 * used to store the fsname and returned to caller.
2605 char *ll_get_fsname(struct super_block *sb, char *buf, int buflen)
2607 static char fsname_static[MTI_NAME_MAXLEN];
2608 struct lustre_sb_info *lsi = s2lsi(sb);
2613 /* this means the caller wants to use static buffer
2614 * and it doesn't care about race. Usually this is
2615 * in error reporting path */
2616 buf = fsname_static;
2617 buflen = sizeof(fsname_static);
2620 len = strlen(lsi->lsi_lmd->lmd_profile);
2621 ptr = strrchr(lsi->lsi_lmd->lmd_profile, '-');
2622 if (ptr && (strcmp(ptr, "-client") == 0))
2625 if (unlikely(len >= buflen))
2627 strncpy(buf, lsi->lsi_lmd->lmd_profile, len);
2633 static char* ll_d_path(struct dentry *dentry, char *buf, int bufsize)
2640 p.mnt = current->fs->root.mnt;
2642 path = d_path(&p, buf, bufsize);
2647 void ll_dirty_page_discard_warn(struct page *page, int ioret)
2649 char *buf, *path = NULL;
2650 struct dentry *dentry = NULL;
2651 struct inode *inode = page->mapping->host;
2653 /* this can be called inside spin lock so use GFP_ATOMIC. */
2654 buf = (char *)__get_free_page(GFP_ATOMIC);
2656 dentry = d_find_alias(page->mapping->host);
2658 path = ll_d_path(dentry, buf, PAGE_SIZE);
2662 "%s: dirty page discard: %s/fid: "DFID"/%s may get corrupted "
2663 "(rc %d)\n", ll_get_fsname(page->mapping->host->i_sb, NULL, 0),
2664 s2lsi(page->mapping->host->i_sb)->lsi_lmd->lmd_dev,
2665 PFID(ll_inode2fid(inode)),
2666 (path && !IS_ERR(path)) ? path : "", ioret);
2672 free_page((unsigned long)buf);
2675 ssize_t ll_copy_user_md(const struct lov_user_md __user *md,
2676 struct lov_user_md **kbuf)
2678 struct lov_user_md lum;
2682 if (copy_from_user(&lum, md, sizeof(lum)))
2685 lum_size = ll_lov_user_md_size(&lum);
2689 OBD_ALLOC(*kbuf, lum_size);
2693 if (copy_from_user(*kbuf, md, lum_size) != 0) {
2694 OBD_FREE(*kbuf, lum_size);
2702 * Compute llite root squash state after a change of root squash
2703 * configuration setting or add/remove of a lnet nid
2705 void ll_compute_rootsquash_state(struct ll_sb_info *sbi)
2707 struct root_squash_info *squash = &sbi->ll_squash;
2710 struct lnet_process_id id;
2712 /* Update norootsquash flag */
2713 down_write(&squash->rsi_sem);
2714 if (list_empty(&squash->rsi_nosquash_nids))
2715 sbi->ll_flags &= ~LL_SBI_NOROOTSQUASH;
2717 /* Do not apply root squash as soon as one of our NIDs is
2718 * in the nosquash_nids list */
2721 while (LNetGetId(i++, &id) != -ENOENT) {
2722 if (LNET_NETTYP(LNET_NIDNET(id.nid)) == LOLND)
2724 if (cfs_match_nid(id.nid, &squash->rsi_nosquash_nids)) {
2730 sbi->ll_flags |= LL_SBI_NOROOTSQUASH;
2732 sbi->ll_flags &= ~LL_SBI_NOROOTSQUASH;
2734 up_write(&squash->rsi_sem);
2738 * Parse linkea content to extract information about a given hardlink
2740 * \param[in] ldata - Initialized linkea data
2741 * \param[in] linkno - Link identifier
2742 * \param[out] parent_fid - The entry's parent FID
2743 * \param[out] ln - Entry name destination buffer
2745 * \retval 0 on success
2746 * \retval Appropriate negative error code on failure
2748 static int ll_linkea_decode(struct linkea_data *ldata, unsigned int linkno,
2749 struct lu_fid *parent_fid, struct lu_name *ln)
2755 rc = linkea_init_with_rec(ldata);
2759 if (linkno >= ldata->ld_leh->leh_reccount)
2760 /* beyond last link */
2763 linkea_first_entry(ldata);
2764 for (idx = 0; ldata->ld_lee != NULL; idx++) {
2765 linkea_entry_unpack(ldata->ld_lee, &ldata->ld_reclen, ln,
2770 linkea_next_entry(ldata);
2780 * Get parent FID and name of an identified link. Operation is performed for
2781 * a given link number, letting the caller iterate over linkno to list one or
2782 * all links of an entry.
2784 * \param[in] file - File descriptor against which to perform the operation
2785 * \param[in,out] arg - User-filled structure containing the linkno to operate
2786 * on and the available size. It is eventually filled with
2787 * the requested information or left untouched on error
2789 * \retval - 0 on success
2790 * \retval - Appropriate negative error code on failure
2792 int ll_getparent(struct file *file, struct getparent __user *arg)
2794 struct inode *inode = file_inode(file);
2795 struct linkea_data *ldata;
2796 struct lu_buf buf = LU_BUF_NULL;
2798 struct lu_fid parent_fid;
2805 if (!cfs_capable(CFS_CAP_DAC_READ_SEARCH) &&
2806 !(ll_i2sbi(inode)->ll_flags & LL_SBI_USER_FID2PATH))
2809 if (get_user(name_size, &arg->gp_name_size))
2812 if (get_user(linkno, &arg->gp_linkno))
2815 if (name_size > PATH_MAX)
2818 OBD_ALLOC(ldata, sizeof(*ldata));
2822 rc = linkea_data_new(ldata, &buf);
2824 GOTO(ldata_free, rc);
2826 #ifdef HAVE_XATTR_HANDLER_FLAGS
2827 rc = ll_xattr_list(inode, XATTR_NAME_LINK, XATTR_TRUSTED_T, buf.lb_buf,
2828 buf.lb_len, OBD_MD_FLXATTR);
2830 rc = ll_getxattr(file_dentry(file), XATTR_NAME_LINK, buf.lb_buf,
2832 #endif /* HAVE_XATTR_HANDLER_FLAGS */
2836 rc = ll_linkea_decode(ldata, linkno, &parent_fid, &ln);
2840 if (ln.ln_namelen >= name_size)
2841 GOTO(lb_free, rc = -EOVERFLOW);
2843 if (copy_to_user(&arg->gp_fid, &parent_fid, sizeof(arg->gp_fid)))
2844 GOTO(lb_free, rc = -EFAULT);
2846 if (copy_to_user(&arg->gp_name, ln.ln_name, ln.ln_namelen))
2847 GOTO(lb_free, rc = -EFAULT);
2849 if (put_user('\0', arg->gp_name + ln.ln_namelen))
2850 GOTO(lb_free, rc = -EFAULT);
2855 OBD_FREE(ldata, sizeof(*ldata));