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.sun.com/software/products/lustre/docs/GPLv2.pdf
20 * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
21 * CA 95054 USA or visit www.sun.com if you need additional information or
27 * Copyright (c) 2003, 2010, Oracle and/or its affiliates. All rights reserved.
28 * Use is subject to license terms.
30 * Copyright (c) 2011, 2014, Intel Corporation.
33 * This file is part of Lustre, http://www.lustre.org/
34 * Lustre is a trademark of Sun Microsystems, Inc.
36 * lustre/llite/llite_lib.c
38 * Lustre Light Super operations
41 #define DEBUG_SUBSYSTEM S_LLITE
43 #include <linux/module.h>
44 #include <linux/statfs.h>
45 #include <linux/types.h>
46 #include <linux/version.h>
49 #include <lustre_ioctl.h>
50 #include <lustre_ha.h>
51 #include <lustre_dlm.h>
52 #include <lprocfs_status.h>
53 #include <lustre_disk.h>
54 #include <lustre_param.h>
55 #include <lustre_log.h>
56 #include <cl_object.h>
57 #include <obd_cksum.h>
58 #include "llite_internal.h"
60 struct kmem_cache *ll_file_data_slab;
62 static struct list_head ll_super_blocks = LIST_HEAD_INIT(ll_super_blocks);
63 static DEFINE_SPINLOCK(ll_sb_lock);
66 #define log2(n) ffz(~(n))
69 static struct ll_sb_info *ll_init_sbi(void)
71 struct ll_sb_info *sbi = NULL;
73 unsigned long lru_page_max;
83 spin_lock_init(&sbi->ll_lock);
84 mutex_init(&sbi->ll_lco.lco_lock);
85 spin_lock_init(&sbi->ll_pp_extent_lock);
86 spin_lock_init(&sbi->ll_process_lock);
87 sbi->ll_rw_stats_on = 0;
90 pages = si.totalram - si.totalhigh;
91 lru_page_max = pages / 2;
93 /* initialize ll_cache data */
94 sbi->ll_cache = cl_cache_init(lru_page_max);
95 if (sbi->ll_cache == NULL) {
96 OBD_FREE(sbi, sizeof(*sbi));
100 sbi->ll_ra_info.ra_max_pages_per_file = min(pages / 32,
101 SBI_DEFAULT_READAHEAD_MAX);
102 sbi->ll_ra_info.ra_max_pages = sbi->ll_ra_info.ra_max_pages_per_file;
103 sbi->ll_ra_info.ra_max_read_ahead_whole_pages =
104 SBI_DEFAULT_READAHEAD_WHOLE_MAX;
105 INIT_LIST_HEAD(&sbi->ll_conn_chain);
106 INIT_LIST_HEAD(&sbi->ll_orphan_dentry_list);
108 ll_generate_random_uuid(uuid);
109 class_uuid_unparse(uuid, &sbi->ll_sb_uuid);
110 CDEBUG(D_CONFIG, "generated uuid: %s\n", sbi->ll_sb_uuid.uuid);
112 spin_lock(&ll_sb_lock);
113 list_add_tail(&sbi->ll_list, &ll_super_blocks);
114 spin_unlock(&ll_sb_lock);
116 sbi->ll_flags |= LL_SBI_VERBOSE;
117 #ifdef ENABLE_CHECKSUM
118 sbi->ll_flags |= LL_SBI_CHECKSUM;
121 #ifdef HAVE_LRU_RESIZE_SUPPORT
122 sbi->ll_flags |= LL_SBI_LRU_RESIZE;
125 for (i = 0; i <= LL_PROCESS_HIST_MAX; i++) {
126 spin_lock_init(&sbi->ll_rw_extents_info.pp_extents[i].
128 spin_lock_init(&sbi->ll_rw_extents_info.pp_extents[i].
132 /* metadata statahead is enabled by default */
133 sbi->ll_sa_max = LL_SA_RPC_DEF;
134 atomic_set(&sbi->ll_sa_total, 0);
135 atomic_set(&sbi->ll_sa_wrong, 0);
136 atomic_set(&sbi->ll_sa_running, 0);
137 atomic_set(&sbi->ll_agl_total, 0);
138 sbi->ll_flags |= LL_SBI_AGL_ENABLED;
141 sbi->ll_squash.rsi_uid = 0;
142 sbi->ll_squash.rsi_gid = 0;
143 INIT_LIST_HEAD(&sbi->ll_squash.rsi_nosquash_nids);
144 init_rwsem(&sbi->ll_squash.rsi_sem);
149 static void ll_free_sbi(struct super_block *sb)
151 struct ll_sb_info *sbi = ll_s2sbi(sb);
155 spin_lock(&ll_sb_lock);
156 list_del(&sbi->ll_list);
157 spin_unlock(&ll_sb_lock);
158 if (!list_empty(&sbi->ll_squash.rsi_nosquash_nids))
159 cfs_free_nidlist(&sbi->ll_squash.rsi_nosquash_nids);
160 if (sbi->ll_cache != NULL) {
161 cl_cache_decref(sbi->ll_cache);
162 sbi->ll_cache = NULL;
164 OBD_FREE(sbi, sizeof(*sbi));
169 static int client_common_fill_super(struct super_block *sb, char *md, char *dt,
170 struct vfsmount *mnt)
172 struct inode *root = NULL;
173 struct ll_sb_info *sbi = ll_s2sbi(sb);
174 struct obd_device *obd;
175 struct obd_capa *oc = NULL;
176 struct obd_statfs *osfs = NULL;
177 struct ptlrpc_request *request = NULL;
178 struct obd_connect_data *data = NULL;
179 struct obd_uuid *uuid;
180 struct md_op_data *op_data;
181 struct lustre_md lmd;
183 int size, err, checksum;
186 obd = class_name2obd(md);
188 CERROR("MD %s: not setup or attached\n", md);
202 /* indicate the features supported by this client */
203 data->ocd_connect_flags = OBD_CONNECT_IBITS | OBD_CONNECT_NODEVOH |
204 OBD_CONNECT_ATTRFID |
205 OBD_CONNECT_VERSION | OBD_CONNECT_BRW_SIZE |
206 OBD_CONNECT_MDS_CAPA | OBD_CONNECT_OSS_CAPA |
207 OBD_CONNECT_CANCELSET | OBD_CONNECT_FID |
208 OBD_CONNECT_AT | OBD_CONNECT_LOV_V3 |
209 OBD_CONNECT_RMT_CLIENT | OBD_CONNECT_VBR |
210 OBD_CONNECT_FULL20 | OBD_CONNECT_64BITHASH|
211 OBD_CONNECT_EINPROGRESS |
212 OBD_CONNECT_JOBSTATS | OBD_CONNECT_LVB_TYPE |
213 OBD_CONNECT_LAYOUTLOCK | OBD_CONNECT_PINGLESS |
214 OBD_CONNECT_MAX_EASIZE |
215 OBD_CONNECT_FLOCK_DEAD |
216 OBD_CONNECT_DISP_STRIPE | OBD_CONNECT_LFSCK |
217 OBD_CONNECT_OPEN_BY_FID |
218 OBD_CONNECT_DIR_STRIPE;
220 #ifdef HAVE_LRU_RESIZE_SUPPORT
221 if (sbi->ll_flags & LL_SBI_LRU_RESIZE)
222 data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
224 #ifdef CONFIG_FS_POSIX_ACL
225 data->ocd_connect_flags |= OBD_CONNECT_ACL | OBD_CONNECT_UMASK;
228 if (OBD_FAIL_CHECK(OBD_FAIL_MDC_LIGHTWEIGHT))
229 /* flag mdc connection as lightweight, only used for test
230 * purpose, use with care */
231 data->ocd_connect_flags |= OBD_CONNECT_LIGHTWEIGHT;
233 data->ocd_ibits_known = MDS_INODELOCK_FULL;
234 data->ocd_version = LUSTRE_VERSION_CODE;
236 if (sb->s_flags & MS_RDONLY)
237 data->ocd_connect_flags |= OBD_CONNECT_RDONLY;
238 if (sbi->ll_flags & LL_SBI_USER_XATTR)
239 data->ocd_connect_flags |= OBD_CONNECT_XATTR;
241 #ifdef HAVE_MS_FLOCK_LOCK
242 /* force vfs to use lustre handler for flock() calls - bug 10743 */
243 sb->s_flags |= MS_FLOCK_LOCK;
245 #ifdef MS_HAS_NEW_AOPS
246 sb->s_flags |= MS_HAS_NEW_AOPS;
249 if (sbi->ll_flags & LL_SBI_FLOCK)
250 sbi->ll_fop = &ll_file_operations_flock;
251 else if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
252 sbi->ll_fop = &ll_file_operations;
254 sbi->ll_fop = &ll_file_operations_noflock;
257 data->ocd_connect_flags |= OBD_CONNECT_REAL;
258 if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
259 data->ocd_connect_flags |= OBD_CONNECT_RMT_CLIENT_FORCE;
261 data->ocd_brw_size = MD_MAX_BRW_SIZE;
263 err = obd_connect(NULL, &sbi->ll_md_exp, obd, &sbi->ll_sb_uuid, data, NULL);
265 LCONSOLE_ERROR_MSG(0x14f, "An MDT (md %s) is performing "
266 "recovery, of which this client is not a "
267 "part. Please wait for recovery to complete,"
268 " abort, or time out.\n", md);
271 CERROR("cannot connect to %s: rc = %d\n", md, err);
275 sbi->ll_md_exp->exp_connect_data = *data;
277 err = obd_fid_init(sbi->ll_md_exp->exp_obd, sbi->ll_md_exp,
278 LUSTRE_SEQ_METADATA);
280 CERROR("%s: Can't init metadata layer FID infrastructure, "
281 "rc = %d\n", sbi->ll_md_exp->exp_obd->obd_name, err);
285 /* For mount, we only need fs info from MDT0, and also in DNE, it
286 * can make sure the client can be mounted as long as MDT0 is
288 err = obd_statfs(NULL, sbi->ll_md_exp, osfs,
289 cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS),
290 OBD_STATFS_FOR_MDT0);
292 GOTO(out_md_fid, err);
294 /* This needs to be after statfs to ensure connect has finished.
295 * Note that "data" does NOT contain the valid connect reply.
296 * If connecting to a 1.8 server there will be no LMV device, so
297 * we can access the MDC export directly and exp_connect_flags will
298 * be non-zero, but if accessing an upgraded 2.1 server it will
299 * have the correct flags filled in.
300 * XXX: fill in the LMV exp_connect_flags from MDC(s). */
301 valid = exp_connect_flags(sbi->ll_md_exp) & CLIENT_CONNECT_MDT_REQD;
302 if (exp_connect_flags(sbi->ll_md_exp) != 0 &&
303 valid != CLIENT_CONNECT_MDT_REQD) {
306 OBD_ALLOC_WAIT(buf, PAGE_CACHE_SIZE);
307 obd_connect_flags2str(buf, PAGE_CACHE_SIZE,
308 valid ^ CLIENT_CONNECT_MDT_REQD, ",");
309 LCONSOLE_ERROR_MSG(0x170, "Server %s does not support "
310 "feature(s) needed for correct operation "
311 "of this client (%s). Please upgrade "
312 "server or downgrade client.\n",
313 sbi->ll_md_exp->exp_obd->obd_name, buf);
314 OBD_FREE(buf, PAGE_CACHE_SIZE);
315 GOTO(out_md_fid, err = -EPROTO);
318 size = sizeof(*data);
319 err = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_CONN_DATA),
320 KEY_CONN_DATA, &size, data, NULL);
322 CERROR("%s: Get connect data failed: rc = %d\n",
323 sbi->ll_md_exp->exp_obd->obd_name, err);
324 GOTO(out_md_fid, err);
327 LASSERT(osfs->os_bsize);
328 sb->s_blocksize = osfs->os_bsize;
329 sb->s_blocksize_bits = log2(osfs->os_bsize);
330 sb->s_magic = LL_SUPER_MAGIC;
331 sb->s_maxbytes = MAX_LFS_FILESIZE;
332 sbi->ll_namelen = osfs->os_namelen;
333 sbi->ll_max_rw_chunk = LL_DEFAULT_MAX_RW_CHUNK;
335 if ((sbi->ll_flags & LL_SBI_USER_XATTR) &&
336 !(data->ocd_connect_flags & OBD_CONNECT_XATTR)) {
337 LCONSOLE_INFO("Disabling user_xattr feature because "
338 "it is not supported on the server\n");
339 sbi->ll_flags &= ~LL_SBI_USER_XATTR;
342 if (data->ocd_connect_flags & OBD_CONNECT_ACL) {
344 sb->s_flags |= MS_POSIXACL;
346 sbi->ll_flags |= LL_SBI_ACL;
348 LCONSOLE_INFO("client wants to enable acl, but mdt not!\n");
350 sb->s_flags &= ~MS_POSIXACL;
352 sbi->ll_flags &= ~LL_SBI_ACL;
355 if (data->ocd_connect_flags & OBD_CONNECT_RMT_CLIENT) {
356 if (!(sbi->ll_flags & LL_SBI_RMT_CLIENT)) {
357 sbi->ll_flags |= LL_SBI_RMT_CLIENT;
358 LCONSOLE_INFO("client is set as remote by default.\n");
361 if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
362 sbi->ll_flags &= ~LL_SBI_RMT_CLIENT;
363 LCONSOLE_INFO("client claims to be remote, but server "
364 "rejected, forced to be local.\n");
368 if (data->ocd_connect_flags & OBD_CONNECT_MDS_CAPA) {
369 LCONSOLE_INFO("client enabled MDS capability!\n");
370 sbi->ll_flags |= LL_SBI_MDS_CAPA;
373 if (data->ocd_connect_flags & OBD_CONNECT_OSS_CAPA) {
374 LCONSOLE_INFO("client enabled OSS capability!\n");
375 sbi->ll_flags |= LL_SBI_OSS_CAPA;
378 if (data->ocd_connect_flags & OBD_CONNECT_64BITHASH)
379 sbi->ll_flags |= LL_SBI_64BIT_HASH;
381 if (data->ocd_connect_flags & OBD_CONNECT_BRW_SIZE)
382 sbi->ll_md_brw_pages = data->ocd_brw_size >> PAGE_CACHE_SHIFT;
384 sbi->ll_md_brw_pages = 1;
386 if (data->ocd_connect_flags & OBD_CONNECT_LAYOUTLOCK)
387 sbi->ll_flags |= LL_SBI_LAYOUT_LOCK;
389 if (data->ocd_ibits_known & MDS_INODELOCK_XATTR) {
390 if (!(data->ocd_connect_flags & OBD_CONNECT_MAX_EASIZE)) {
391 LCONSOLE_INFO("%s: disabling xattr cache due to "
392 "unknown maximum xattr size.\n", dt);
394 sbi->ll_flags |= LL_SBI_XATTR_CACHE;
395 sbi->ll_xattr_cache_enabled = 1;
399 obd = class_name2obd(dt);
401 CERROR("DT %s: not setup or attached\n", dt);
402 GOTO(out_md_fid, err = -ENODEV);
405 data->ocd_connect_flags = OBD_CONNECT_GRANT | OBD_CONNECT_VERSION |
406 OBD_CONNECT_REQPORTAL | OBD_CONNECT_BRW_SIZE |
407 OBD_CONNECT_CANCELSET | OBD_CONNECT_FID |
408 OBD_CONNECT_SRVLOCK | OBD_CONNECT_TRUNCLOCK|
409 OBD_CONNECT_AT | OBD_CONNECT_RMT_CLIENT |
410 OBD_CONNECT_OSS_CAPA | OBD_CONNECT_VBR|
411 OBD_CONNECT_FULL20 | OBD_CONNECT_64BITHASH |
412 OBD_CONNECT_MAXBYTES |
413 OBD_CONNECT_EINPROGRESS |
414 OBD_CONNECT_JOBSTATS | OBD_CONNECT_LVB_TYPE |
415 OBD_CONNECT_LAYOUTLOCK |
416 OBD_CONNECT_PINGLESS | OBD_CONNECT_LFSCK;
418 if (!OBD_FAIL_CHECK(OBD_FAIL_OSC_CONNECT_CKSUM)) {
419 /* OBD_CONNECT_CKSUM should always be set, even if checksums are
420 * disabled by default, because it can still be enabled on the
421 * fly via /proc. As a consequence, we still need to come to an
422 * agreement on the supported algorithms at connect time */
423 data->ocd_connect_flags |= OBD_CONNECT_CKSUM;
425 if (OBD_FAIL_CHECK(OBD_FAIL_OSC_CKSUM_ADLER_ONLY))
426 data->ocd_cksum_types = OBD_CKSUM_ADLER;
428 data->ocd_cksum_types = cksum_types_supported_client();
431 #ifdef HAVE_LRU_RESIZE_SUPPORT
432 data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
434 if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
435 data->ocd_connect_flags |= OBD_CONNECT_RMT_CLIENT_FORCE;
437 CDEBUG(D_RPCTRACE, "ocd_connect_flags: "LPX64" ocd_version: %d "
438 "ocd_grant: %d\n", data->ocd_connect_flags,
439 data->ocd_version, data->ocd_grant);
441 obd->obd_upcall.onu_owner = &sbi->ll_lco;
442 obd->obd_upcall.onu_upcall = cl_ocd_update;
444 data->ocd_brw_size = DT_MAX_BRW_SIZE;
446 err = obd_connect(NULL, &sbi->ll_dt_exp, obd, &sbi->ll_sb_uuid, data,
449 LCONSOLE_ERROR_MSG(0x150, "An OST (dt %s) is performing "
450 "recovery, of which this client is not a "
451 "part. Please wait for recovery to "
452 "complete, abort, or time out.\n", dt);
455 CERROR("%s: Cannot connect to %s: rc = %d\n",
456 sbi->ll_dt_exp->exp_obd->obd_name, dt, err);
460 sbi->ll_dt_exp->exp_connect_data = *data;
462 err = obd_fid_init(sbi->ll_dt_exp->exp_obd, sbi->ll_dt_exp,
463 LUSTRE_SEQ_METADATA);
465 CERROR("%s: Can't init data layer FID infrastructure, "
466 "rc = %d\n", sbi->ll_dt_exp->exp_obd->obd_name, err);
470 mutex_lock(&sbi->ll_lco.lco_lock);
471 sbi->ll_lco.lco_flags = data->ocd_connect_flags;
472 sbi->ll_lco.lco_md_exp = sbi->ll_md_exp;
473 sbi->ll_lco.lco_dt_exp = sbi->ll_dt_exp;
474 mutex_unlock(&sbi->ll_lco.lco_lock);
476 fid_zero(&sbi->ll_root_fid);
477 err = md_getstatus(sbi->ll_md_exp, &sbi->ll_root_fid, &oc);
479 CERROR("cannot mds_connect: rc = %d\n", err);
480 GOTO(out_lock_cn_cb, err);
482 if (!fid_is_sane(&sbi->ll_root_fid)) {
483 CERROR("%s: Invalid root fid "DFID" during mount\n",
484 sbi->ll_md_exp->exp_obd->obd_name,
485 PFID(&sbi->ll_root_fid));
486 GOTO(out_lock_cn_cb, err = -EINVAL);
488 CDEBUG(D_SUPER, "rootfid "DFID"\n", PFID(&sbi->ll_root_fid));
490 sb->s_op = &lustre_super_operations;
491 #if THREAD_SIZE >= 8192 /*b=17630*/
492 sb->s_export_op = &lustre_export_operations;
496 * XXX: move this to after cbd setup? */
497 valid = OBD_MD_FLGETATTR | OBD_MD_FLBLOCKS | OBD_MD_FLMDSCAPA |
499 if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
500 valid |= OBD_MD_FLRMTPERM;
501 else if (sbi->ll_flags & LL_SBI_ACL)
502 valid |= OBD_MD_FLACL;
504 OBD_ALLOC_PTR(op_data);
506 GOTO(out_lock_cn_cb, err = -ENOMEM);
508 op_data->op_fid1 = sbi->ll_root_fid;
509 op_data->op_mode = 0;
510 op_data->op_capa1 = oc;
511 op_data->op_valid = valid;
513 err = md_getattr(sbi->ll_md_exp, op_data, &request);
516 OBD_FREE_PTR(op_data);
518 CERROR("%s: md_getattr failed for root: rc = %d\n",
519 sbi->ll_md_exp->exp_obd->obd_name, err);
520 GOTO(out_lock_cn_cb, err);
523 err = md_get_lustre_md(sbi->ll_md_exp, request, sbi->ll_dt_exp,
524 sbi->ll_md_exp, &lmd);
526 CERROR("failed to understand root inode md: rc = %d\n", err);
527 ptlrpc_req_finished(request);
528 GOTO(out_lock_cn_cb, err);
531 LASSERT(fid_is_sane(&sbi->ll_root_fid));
532 root = ll_iget(sb, cl_fid_build_ino(&sbi->ll_root_fid,
533 sbi->ll_flags & LL_SBI_32BIT_API),
535 md_free_lustre_md(sbi->ll_md_exp, &lmd);
536 ptlrpc_req_finished(request);
540 obd_free_memmd(sbi->ll_dt_exp, &lmd.lsm);
541 #ifdef CONFIG_FS_POSIX_ACL
543 posix_acl_release(lmd.posix_acl);
544 lmd.posix_acl = NULL;
547 err = IS_ERR(root) ? PTR_ERR(root) : -EBADF;
549 CERROR("lustre_lite: bad iget4 for root\n");
553 #ifdef CONFIG_FS_POSIX_ACL
554 if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
555 rct_init(&sbi->ll_rct);
556 et_init(&sbi->ll_et);
560 checksum = sbi->ll_flags & LL_SBI_CHECKSUM;
561 err = obd_set_info_async(NULL, sbi->ll_dt_exp, sizeof(KEY_CHECKSUM),
562 KEY_CHECKSUM, sizeof(checksum), &checksum,
566 err = obd_set_info_async(NULL, sbi->ll_dt_exp, sizeof(KEY_CACHE_SET),
567 KEY_CACHE_SET, sizeof(*sbi->ll_cache),
568 sbi->ll_cache, NULL);
570 sb->s_root = d_make_root(root);
571 if (sb->s_root == NULL) {
572 CERROR("%s: can't make root dentry\n",
573 ll_get_fsname(sb, NULL, 0));
574 GOTO(out_root, err = -ENOMEM);
576 #ifdef HAVE_DCACHE_LOCK
577 sb->s_root->d_op = &ll_d_ops;
580 sbi->ll_sdev_orig = sb->s_dev;
582 /* We set sb->s_dev equal on all lustre clients in order to support
583 * NFS export clustering. NFSD requires that the FSID be the same
585 /* s_dev is also used in lt_compare() to compare two fs, but that is
586 * only a node-local comparison. */
587 uuid = obd_get_uuid(sbi->ll_md_exp);
589 sb->s_dev = get_uuid2int(uuid->uuid, strlen(uuid->uuid));
595 if (proc_lustre_fs_root != NULL) {
596 err = lprocfs_register_mountpoint(proc_lustre_fs_root, sb,
599 CERROR("%s: could not register mount in lprocfs: "
600 "rc = %d\n", ll_get_fsname(sb, NULL, 0), err);
610 obd_fid_fini(sbi->ll_dt_exp->exp_obd);
612 obd_disconnect(sbi->ll_dt_exp);
613 sbi->ll_dt_exp = NULL;
615 obd_fid_fini(sbi->ll_md_exp->exp_obd);
617 obd_disconnect(sbi->ll_md_exp);
618 sbi->ll_md_exp = NULL;
627 int ll_get_max_mdsize(struct ll_sb_info *sbi, int *lmmsize)
631 *lmmsize = obd_size_diskmd(sbi->ll_dt_exp, NULL);
633 rc = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_MAX_EASIZE),
634 KEY_MAX_EASIZE, &size, lmmsize, NULL);
636 CERROR("Get max mdsize error rc %d\n", rc);
642 * Get the value of the default_easize parameter.
644 * \see client_obd::cl_default_mds_easize
646 * \param[in] sbi superblock info for this filesystem
647 * \param[out] lmmsize pointer to storage location for value
649 * \retval 0 on success
650 * \retval negative negated errno on failure
652 int ll_get_default_mdsize(struct ll_sb_info *sbi, int *lmmsize)
657 rc = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_DEFAULT_EASIZE),
658 KEY_DEFAULT_EASIZE, &size, lmmsize, NULL);
660 CERROR("Get default mdsize error rc %d\n", rc);
666 * Set the default_easize parameter to the given value.
668 * \see client_obd::cl_default_mds_easize
670 * \param[in] sbi superblock info for this filesystem
671 * \param[in] lmmsize the size to set
673 * \retval 0 on success
674 * \retval negative negated errno on failure
676 int ll_set_default_mdsize(struct ll_sb_info *sbi, int lmmsize)
680 if (lmmsize < sizeof(struct lov_mds_md) ||
681 lmmsize > OBD_MAX_DEFAULT_EA_SIZE)
684 rc = obd_set_info_async(NULL, sbi->ll_md_exp,
685 sizeof(KEY_DEFAULT_EASIZE), KEY_DEFAULT_EASIZE,
686 sizeof(int), &lmmsize, NULL);
691 int ll_get_max_cookiesize(struct ll_sb_info *sbi, int *lmmsize)
696 rc = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_MAX_COOKIESIZE),
697 KEY_MAX_COOKIESIZE, &size, lmmsize, NULL);
699 CERROR("Get max cookiesize error rc %d\n", rc);
704 int ll_get_default_cookiesize(struct ll_sb_info *sbi, int *lmmsize)
709 rc = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_DEFAULT_COOKIESIZE),
710 KEY_DEFAULT_COOKIESIZE, &size, lmmsize, NULL);
712 CERROR("Get default cookiesize error rc %d\n", rc);
717 static void ll_dump_inode(struct inode *inode)
719 struct ll_d_hlist_node *tmp;
720 int dentry_count = 0;
722 LASSERT(inode != NULL);
724 ll_d_hlist_for_each(tmp, &inode->i_dentry)
727 CERROR("%s: inode %p dump: dev=%s fid="DFID
728 " mode=%o count=%u, %d dentries\n",
729 ll_get_fsname(inode->i_sb, NULL, 0), inode,
730 ll_i2mdexp(inode)->exp_obd->obd_name, PFID(ll_inode2fid(inode)),
731 inode->i_mode, atomic_read(&inode->i_count), dentry_count);
734 void lustre_dump_dentry(struct dentry *dentry, int recur)
736 struct list_head *tmp;
739 LASSERT(dentry != NULL);
741 list_for_each(tmp, &dentry->d_subdirs)
744 CERROR("dentry %p dump: name=%.*s parent=%.*s (%p), inode=%p, count=%u,"
745 " flags=0x%x, fsdata=%p, %d subdirs\n", dentry,
746 dentry->d_name.len, dentry->d_name.name,
747 dentry->d_parent->d_name.len, dentry->d_parent->d_name.name,
748 dentry->d_parent, dentry->d_inode, ll_d_count(dentry),
749 dentry->d_flags, dentry->d_fsdata, subdirs);
750 if (dentry->d_inode != NULL)
751 ll_dump_inode(dentry->d_inode);
756 list_for_each(tmp, &dentry->d_subdirs) {
757 struct dentry *d = list_entry(tmp, struct dentry, d_u.d_child);
758 lustre_dump_dentry(d, recur - 1);
762 static void client_common_put_super(struct super_block *sb)
764 struct ll_sb_info *sbi = ll_s2sbi(sb);
767 #ifdef CONFIG_FS_POSIX_ACL
768 if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
769 et_fini(&sbi->ll_et);
770 rct_fini(&sbi->ll_rct);
776 list_del(&sbi->ll_conn_chain);
778 obd_fid_fini(sbi->ll_dt_exp->exp_obd);
779 obd_disconnect(sbi->ll_dt_exp);
780 sbi->ll_dt_exp = NULL;
782 lprocfs_unregister_mountpoint(sbi);
784 obd_fid_fini(sbi->ll_md_exp->exp_obd);
785 obd_disconnect(sbi->ll_md_exp);
786 sbi->ll_md_exp = NULL;
791 void ll_kill_super(struct super_block *sb)
793 struct ll_sb_info *sbi;
797 if (!(sb->s_flags & MS_ACTIVE))
801 /* we need restore s_dev from changed for clustred NFS before put_super
802 * because new kernels have cached s_dev and change sb->s_dev in
803 * put_super not affected real removing devices */
805 sb->s_dev = sbi->ll_sdev_orig;
806 sbi->ll_umounting = 1;
808 /* wait running statahead threads to quit */
809 while (atomic_read(&sbi->ll_sa_running) > 0)
810 schedule_timeout_and_set_state(TASK_UNINTERRUPTIBLE,
811 msecs_to_jiffies(MSEC_PER_SEC >> 3));
817 static inline int ll_set_opt(const char *opt, char *data, int fl)
819 if (strncmp(opt, data, strlen(opt)) != 0)
825 /* non-client-specific mount options are parsed in lmd_parse */
826 static int ll_options(char *options, int *flags)
829 char *s1 = options, *s2;
835 CDEBUG(D_CONFIG, "Parsing opts %s\n", options);
838 CDEBUG(D_SUPER, "next opt=%s\n", s1);
839 tmp = ll_set_opt("nolock", s1, LL_SBI_NOLCK);
844 tmp = ll_set_opt("flock", s1, LL_SBI_FLOCK);
849 tmp = ll_set_opt("localflock", s1, LL_SBI_LOCALFLOCK);
854 tmp = ll_set_opt("noflock", s1, LL_SBI_FLOCK|LL_SBI_LOCALFLOCK);
859 tmp = ll_set_opt("user_xattr", s1, LL_SBI_USER_XATTR);
864 tmp = ll_set_opt("nouser_xattr", s1, LL_SBI_USER_XATTR);
869 tmp = ll_set_opt("remote_client", s1, LL_SBI_RMT_CLIENT);
874 tmp = ll_set_opt("user_fid2path", s1, LL_SBI_USER_FID2PATH);
879 tmp = ll_set_opt("nouser_fid2path", s1, LL_SBI_USER_FID2PATH);
885 tmp = ll_set_opt("checksum", s1, LL_SBI_CHECKSUM);
890 tmp = ll_set_opt("nochecksum", s1, LL_SBI_CHECKSUM);
895 tmp = ll_set_opt("lruresize", s1, LL_SBI_LRU_RESIZE);
900 tmp = ll_set_opt("nolruresize", s1, LL_SBI_LRU_RESIZE);
905 tmp = ll_set_opt("lazystatfs", s1, LL_SBI_LAZYSTATFS);
910 tmp = ll_set_opt("nolazystatfs", s1, LL_SBI_LAZYSTATFS);
915 tmp = ll_set_opt("32bitapi", s1, LL_SBI_32BIT_API);
920 tmp = ll_set_opt("verbose", s1, LL_SBI_VERBOSE);
925 tmp = ll_set_opt("noverbose", s1, LL_SBI_VERBOSE);
930 LCONSOLE_ERROR_MSG(0x152, "Unknown option '%s', won't mount.\n",
936 s2 = strchr(s1, ',');
944 void ll_lli_init(struct ll_inode_info *lli)
946 lli->lli_inode_magic = LLI_INODE_MAGIC;
948 lli->lli_maxbytes = MAX_LFS_FILESIZE;
949 spin_lock_init(&lli->lli_lock);
950 lli->lli_posix_acl = NULL;
951 lli->lli_remote_perms = NULL;
952 mutex_init(&lli->lli_rmtperm_mutex);
953 /* Do not set lli_fid, it has been initialized already. */
954 fid_zero(&lli->lli_pfid);
955 INIT_LIST_HEAD(&lli->lli_oss_capas);
956 atomic_set(&lli->lli_open_count, 0);
957 lli->lli_mds_capa = NULL;
958 lli->lli_rmtperm_time = 0;
959 lli->lli_mds_read_och = NULL;
960 lli->lli_mds_write_och = NULL;
961 lli->lli_mds_exec_och = NULL;
962 lli->lli_open_fd_read_count = 0;
963 lli->lli_open_fd_write_count = 0;
964 lli->lli_open_fd_exec_count = 0;
965 mutex_init(&lli->lli_och_mutex);
966 spin_lock_init(&lli->lli_agl_lock);
967 lli->lli_has_smd = false;
968 spin_lock_init(&lli->lli_layout_lock);
969 ll_layout_version_set(lli, LL_LAYOUT_GEN_NONE);
970 lli->lli_clob = NULL;
972 init_rwsem(&lli->lli_xattrs_list_rwsem);
973 mutex_init(&lli->lli_xattrs_enq_lock);
975 LASSERT(lli->lli_vfs_inode.i_mode != 0);
976 if (S_ISDIR(lli->lli_vfs_inode.i_mode)) {
977 mutex_init(&lli->lli_readdir_mutex);
978 lli->lli_opendir_key = NULL;
980 spin_lock_init(&lli->lli_sa_lock);
981 lli->lli_opendir_pid = 0;
982 lli->lli_sa_enabled = 0;
983 lli->lli_def_stripe_offset = -1;
985 mutex_init(&lli->lli_size_mutex);
986 lli->lli_symlink_name = NULL;
987 init_rwsem(&lli->lli_trunc_sem);
988 range_lock_tree_init(&lli->lli_write_tree);
989 init_rwsem(&lli->lli_glimpse_sem);
990 lli->lli_glimpse_time = 0;
991 INIT_LIST_HEAD(&lli->lli_agl_list);
992 lli->lli_agl_index = 0;
993 lli->lli_async_rc = 0;
995 mutex_init(&lli->lli_layout_mutex);
998 static inline int ll_bdi_register(struct backing_dev_info *bdi)
1000 static atomic_t ll_bdi_num = ATOMIC_INIT(0);
1002 bdi->name = "lustre";
1003 return bdi_register(bdi, NULL, "lustre-%d",
1004 atomic_inc_return(&ll_bdi_num));
1007 int ll_fill_super(struct super_block *sb, struct vfsmount *mnt)
1009 struct lustre_profile *lprof = NULL;
1010 struct lustre_sb_info *lsi = s2lsi(sb);
1011 struct ll_sb_info *sbi;
1012 char *dt = NULL, *md = NULL;
1013 char *profilenm = get_profile_name(sb);
1014 struct config_llog_instance *cfg;
1015 /* %p for void* in printf needs 16+2 characters: 0xffffffffffffffff */
1016 const int instlen = sizeof(cfg->cfg_instance) * 2 + 2;
1020 CDEBUG(D_VFSTRACE, "VFS Op: sb %p\n", sb);
1026 try_module_get(THIS_MODULE);
1028 /* client additional sb info */
1029 lsi->lsi_llsbi = sbi = ll_init_sbi();
1031 module_put(THIS_MODULE);
1036 err = ll_options(lsi->lsi_lmd->lmd_opts, &sbi->ll_flags);
1038 GOTO(out_free, err);
1040 err = bdi_init(&lsi->lsi_bdi);
1042 GOTO(out_free, err);
1043 lsi->lsi_flags |= LSI_BDI_INITIALIZED;
1044 lsi->lsi_bdi.capabilities = BDI_CAP_MAP_COPY;
1045 err = ll_bdi_register(&lsi->lsi_bdi);
1047 GOTO(out_free, err);
1049 sb->s_bdi = &lsi->lsi_bdi;
1050 #ifndef HAVE_DCACHE_LOCK
1051 /* kernel >= 2.6.38 store dentry operations in sb->s_d_op. */
1052 sb->s_d_op = &ll_d_ops;
1055 /* Generate a string unique to this super, in case some joker tries
1056 to mount the same fs at two mount points.
1057 Use the address of the super itself.*/
1058 cfg->cfg_instance = sb;
1059 cfg->cfg_uuid = lsi->lsi_llsbi->ll_sb_uuid;
1060 cfg->cfg_callback = class_config_llog_handler;
1061 /* set up client obds */
1062 err = lustre_process_log(sb, profilenm, cfg);
1064 GOTO(out_free, err);
1066 /* Profile set with LCFG_MOUNTOPT so we can find our mdc and osc obds */
1067 lprof = class_get_profile(profilenm);
1068 if (lprof == NULL) {
1069 LCONSOLE_ERROR_MSG(0x156, "The client profile '%s' could not be"
1070 " read from the MGS. Does that filesystem "
1071 "exist?\n", profilenm);
1072 GOTO(out_free, err = -EINVAL);
1074 CDEBUG(D_CONFIG, "Found profile %s: mdc=%s osc=%s\n", profilenm,
1075 lprof->lp_md, lprof->lp_dt);
1077 OBD_ALLOC(dt, strlen(lprof->lp_dt) + instlen + 2);
1079 GOTO(out_free, err = -ENOMEM);
1080 sprintf(dt, "%s-%p", lprof->lp_dt, cfg->cfg_instance);
1082 OBD_ALLOC(md, strlen(lprof->lp_md) + instlen + 2);
1084 GOTO(out_free, err = -ENOMEM);
1085 sprintf(md, "%s-%p", lprof->lp_md, cfg->cfg_instance);
1087 /* connections, registrations, sb setup */
1088 err = client_common_fill_super(sb, md, dt, mnt);
1090 GOTO(out_free, err);
1092 sbi->ll_client_common_fill_super_succeeded = 1;
1096 OBD_FREE(md, strlen(lprof->lp_md) + instlen + 2);
1098 OBD_FREE(dt, strlen(lprof->lp_dt) + instlen + 2);
1101 else if (sbi->ll_flags & LL_SBI_VERBOSE)
1102 LCONSOLE_WARN("Mounted %s\n", profilenm);
1106 } /* ll_fill_super */
1108 void ll_put_super(struct super_block *sb)
1110 struct config_llog_instance cfg, params_cfg;
1111 struct obd_device *obd;
1112 struct lustre_sb_info *lsi = s2lsi(sb);
1113 struct ll_sb_info *sbi = ll_s2sbi(sb);
1114 char *profilenm = get_profile_name(sb);
1116 int next, force = 1, rc = 0;
1119 CDEBUG(D_VFSTRACE, "VFS Op: sb %p - %s\n", sb, profilenm);
1121 ll_print_capa_stat(sbi);
1123 cfg.cfg_instance = sb;
1124 lustre_end_log(sb, profilenm, &cfg);
1126 params_cfg.cfg_instance = sb;
1127 lustre_end_log(sb, PARAMS_FILENAME, ¶ms_cfg);
1129 if (sbi->ll_md_exp) {
1130 obd = class_exp2obd(sbi->ll_md_exp);
1132 force = obd->obd_force;
1135 /* Wait for unstable pages to be committed to stable storage */
1137 struct l_wait_info lwi = LWI_INTR(LWI_ON_SIGNAL_NOOP, NULL);
1138 rc = l_wait_event(sbi->ll_cache->ccc_unstable_waitq,
1139 atomic_long_read(&sbi->ll_cache->ccc_unstable_nr) == 0,
1143 ccc_count = atomic_long_read(&sbi->ll_cache->ccc_unstable_nr);
1144 if (force == 0 && rc != -EINTR)
1145 LASSERTF(ccc_count == 0, "count: %li\n", ccc_count);
1148 /* We need to set force before the lov_disconnect in
1149 lustre_common_put_super, since l_d cleans up osc's as well. */
1152 while ((obd = class_devices_in_group(&sbi->ll_sb_uuid,
1154 obd->obd_force = force;
1158 if (sbi->ll_client_common_fill_super_succeeded) {
1159 /* Only if client_common_fill_super succeeded */
1160 client_common_put_super(sb);
1164 while ((obd = class_devices_in_group(&sbi->ll_sb_uuid, &next)) !=NULL) {
1165 class_manual_cleanup(obd);
1168 if (sbi->ll_flags & LL_SBI_VERBOSE)
1169 LCONSOLE_WARN("Unmounted %s\n", profilenm ? profilenm : "");
1172 class_del_profile(profilenm);
1174 if (lsi->lsi_flags & LSI_BDI_INITIALIZED) {
1175 bdi_destroy(&lsi->lsi_bdi);
1176 lsi->lsi_flags &= ~LSI_BDI_INITIALIZED;
1180 lsi->lsi_llsbi = NULL;
1182 lustre_common_put_super(sb);
1184 cl_env_cache_purge(~0);
1186 module_put(THIS_MODULE);
1189 } /* client_put_super */
1191 struct inode *ll_inode_from_resource_lock(struct ldlm_lock *lock)
1193 struct inode *inode = NULL;
1195 /* NOTE: we depend on atomic igrab() -bzzz */
1196 lock_res_and_lock(lock);
1197 if (lock->l_resource->lr_lvb_inode) {
1198 struct ll_inode_info * lli;
1199 lli = ll_i2info(lock->l_resource->lr_lvb_inode);
1200 if (lli->lli_inode_magic == LLI_INODE_MAGIC) {
1201 inode = igrab(lock->l_resource->lr_lvb_inode);
1203 inode = lock->l_resource->lr_lvb_inode;
1204 LDLM_DEBUG_LIMIT(inode->i_state & I_FREEING ? D_INFO :
1205 D_WARNING, lock, "lr_lvb_inode %p is "
1206 "bogus: magic %08x",
1207 lock->l_resource->lr_lvb_inode,
1208 lli->lli_inode_magic);
1212 unlock_res_and_lock(lock);
1216 static void ll_dir_clear_lsm_md(struct inode *inode)
1218 struct ll_inode_info *lli = ll_i2info(inode);
1220 LASSERT(S_ISDIR(inode->i_mode));
1222 if (lli->lli_lsm_md != NULL) {
1223 lmv_free_memmd(lli->lli_lsm_md);
1224 lli->lli_lsm_md = NULL;
1228 static struct inode *ll_iget_anon_dir(struct super_block *sb,
1229 const struct lu_fid *fid,
1230 struct lustre_md *md)
1232 struct ll_sb_info *sbi = ll_s2sbi(sb);
1233 struct mdt_body *body = md->body;
1234 struct inode *inode;
1238 ino = cl_fid_build_ino(fid, sbi->ll_flags & LL_SBI_32BIT_API);
1239 inode = iget_locked(sb, ino);
1240 if (inode == NULL) {
1241 CERROR("%s: failed get simple inode "DFID": rc = -ENOENT\n",
1242 ll_get_fsname(sb, NULL, 0), PFID(fid));
1243 RETURN(ERR_PTR(-ENOENT));
1246 if (inode->i_state & I_NEW) {
1247 struct ll_inode_info *lli = ll_i2info(inode);
1248 struct lmv_stripe_md *lsm = md->lmv;
1250 inode->i_mode = (inode->i_mode & ~S_IFMT) |
1251 (body->mbo_mode & S_IFMT);
1252 LASSERTF(S_ISDIR(inode->i_mode), "Not slave inode "DFID"\n",
1255 LTIME_S(inode->i_mtime) = 0;
1256 LTIME_S(inode->i_atime) = 0;
1257 LTIME_S(inode->i_ctime) = 0;
1260 /* initializing backing dev info. */
1261 inode->i_mapping->backing_dev_info =
1262 &s2lsi(inode->i_sb)->lsi_bdi;
1263 inode->i_op = &ll_dir_inode_operations;
1264 inode->i_fop = &ll_dir_operations;
1265 lli->lli_fid = *fid;
1268 LASSERT(lsm != NULL);
1269 /* master object FID */
1270 lli->lli_pfid = body->mbo_fid1;
1271 CDEBUG(D_INODE, "lli %p slave "DFID" master "DFID"\n",
1272 lli, PFID(fid), PFID(&lli->lli_pfid));
1273 unlock_new_inode(inode);
1279 static int ll_init_lsm_md(struct inode *inode, struct lustre_md *md)
1282 struct lmv_stripe_md *lsm = md->lmv;
1285 LASSERT(lsm != NULL);
1286 /* XXX sigh, this lsm_root initialization should be in
1287 * LMV layer, but it needs ll_iget right now, so we
1288 * put this here right now. */
1289 for (i = 0; i < lsm->lsm_md_stripe_count; i++) {
1290 fid = &lsm->lsm_md_oinfo[i].lmo_fid;
1291 LASSERT(lsm->lsm_md_oinfo[i].lmo_root == NULL);
1292 /* Unfortunately ll_iget will call ll_update_inode,
1293 * where the initialization of slave inode is slightly
1294 * different, so it reset lsm_md to NULL to avoid
1295 * initializing lsm for slave inode. */
1296 /* For migrating inode, master stripe and master object will
1297 * be same, so we only need assign this inode */
1298 if (lsm->lsm_md_hash_type & LMV_HASH_FLAG_MIGRATION && i == 0)
1299 lsm->lsm_md_oinfo[i].lmo_root = inode;
1301 lsm->lsm_md_oinfo[i].lmo_root =
1302 ll_iget_anon_dir(inode->i_sb, fid, md);
1304 if (IS_ERR(lsm->lsm_md_oinfo[i].lmo_root)) {
1305 int rc = PTR_ERR(lsm->lsm_md_oinfo[i].lmo_root);
1307 lsm->lsm_md_oinfo[i].lmo_root = NULL;
1315 static inline int lli_lsm_md_eq(const struct lmv_stripe_md *lsm_md1,
1316 const struct lmv_stripe_md *lsm_md2)
1318 return lsm_md1->lsm_md_magic == lsm_md2->lsm_md_magic &&
1319 lsm_md1->lsm_md_stripe_count == lsm_md2->lsm_md_stripe_count &&
1320 lsm_md1->lsm_md_master_mdt_index ==
1321 lsm_md2->lsm_md_master_mdt_index &&
1322 lsm_md1->lsm_md_hash_type == lsm_md2->lsm_md_hash_type &&
1323 lsm_md1->lsm_md_layout_version ==
1324 lsm_md2->lsm_md_layout_version &&
1325 strcmp(lsm_md1->lsm_md_pool_name,
1326 lsm_md2->lsm_md_pool_name) == 0;
1329 static int ll_update_lsm_md(struct inode *inode, struct lustre_md *md)
1331 struct ll_inode_info *lli = ll_i2info(inode);
1332 struct lmv_stripe_md *lsm = md->lmv;
1336 LASSERT(S_ISDIR(inode->i_mode));
1337 CDEBUG(D_INODE, "update lsm %p of "DFID"\n", lli->lli_lsm_md,
1338 PFID(ll_inode2fid(inode)));
1340 /* no striped information from request. */
1342 if (lli->lli_lsm_md == NULL) {
1344 } else if (lli->lli_lsm_md->lsm_md_hash_type &
1345 LMV_HASH_FLAG_MIGRATION) {
1346 /* migration is done, the temporay MIGRATE layout has
1348 CDEBUG(D_INODE, DFID" finish migration.\n",
1349 PFID(ll_inode2fid(inode)));
1350 lmv_free_memmd(lli->lli_lsm_md);
1351 lli->lli_lsm_md = NULL;
1354 /* The lustre_md from req does not include stripeEA,
1355 * see ll_md_setattr */
1360 /* set the directory layout */
1361 if (lli->lli_lsm_md == NULL) {
1363 rc = ll_init_lsm_md(inode, md);
1367 lli->lli_lsm_md = lsm;
1368 /* set lsm_md to NULL, so the following free lustre_md
1369 * will not free this lsm */
1371 CDEBUG(D_INODE, "Set lsm %p magic %x to "DFID"\n", lsm,
1372 lsm->lsm_md_magic, PFID(ll_inode2fid(inode)));
1376 /* Compare the old and new stripe information */
1377 if (!lsm_md_eq(lli->lli_lsm_md, lsm)) {
1378 struct lmv_stripe_md *old_lsm = lli->lli_lsm_md;
1381 CERROR("%s: inode "DFID"(%p)'s lmv layout mismatch (%p)/(%p)"
1382 "magic:0x%x/0x%x stripe count: %d/%d master_mdt: %d/%d"
1383 "hash_type:0x%x/0x%x layout: 0x%x/0x%x pool:%s/%s\n",
1384 ll_get_fsname(inode->i_sb, NULL, 0), PFID(&lli->lli_fid),
1385 inode, lsm, old_lsm,
1386 lsm->lsm_md_magic, old_lsm->lsm_md_magic,
1387 lsm->lsm_md_stripe_count,
1388 old_lsm->lsm_md_stripe_count,
1389 lsm->lsm_md_master_mdt_index,
1390 old_lsm->lsm_md_master_mdt_index,
1391 lsm->lsm_md_hash_type, old_lsm->lsm_md_hash_type,
1392 lsm->lsm_md_layout_version,
1393 old_lsm->lsm_md_layout_version,
1394 lsm->lsm_md_pool_name,
1395 old_lsm->lsm_md_pool_name);
1397 for (idx = 0; idx < old_lsm->lsm_md_stripe_count; idx++) {
1398 CERROR("%s: sub FIDs in old lsm idx %d, old: "DFID"\n",
1399 ll_get_fsname(inode->i_sb, NULL, 0), idx,
1400 PFID(&old_lsm->lsm_md_oinfo[idx].lmo_fid));
1403 for (idx = 0; idx < lsm->lsm_md_stripe_count; idx++) {
1404 CERROR("%s: sub FIDs in new lsm idx %d, new: "DFID"\n",
1405 ll_get_fsname(inode->i_sb, NULL, 0), idx,
1406 PFID(&lsm->lsm_md_oinfo[idx].lmo_fid));
1415 void ll_clear_inode(struct inode *inode)
1417 struct ll_inode_info *lli = ll_i2info(inode);
1418 struct ll_sb_info *sbi = ll_i2sbi(inode);
1421 CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p)\n",
1422 PFID(ll_inode2fid(inode)), inode);
1424 if (S_ISDIR(inode->i_mode)) {
1425 /* these should have been cleared in ll_file_release */
1426 LASSERT(lli->lli_opendir_key == NULL);
1427 LASSERT(lli->lli_sai == NULL);
1428 LASSERT(lli->lli_opendir_pid == 0);
1431 md_null_inode(sbi->ll_md_exp, ll_inode2fid(inode));
1433 LASSERT(!lli->lli_open_fd_write_count);
1434 LASSERT(!lli->lli_open_fd_read_count);
1435 LASSERT(!lli->lli_open_fd_exec_count);
1437 if (lli->lli_mds_write_och)
1438 ll_md_real_close(inode, FMODE_WRITE);
1439 if (lli->lli_mds_exec_och)
1440 ll_md_real_close(inode, FMODE_EXEC);
1441 if (lli->lli_mds_read_och)
1442 ll_md_real_close(inode, FMODE_READ);
1444 if (S_ISLNK(inode->i_mode) && lli->lli_symlink_name) {
1445 OBD_FREE(lli->lli_symlink_name,
1446 strlen(lli->lli_symlink_name) + 1);
1447 lli->lli_symlink_name = NULL;
1450 ll_xattr_cache_destroy(inode);
1452 if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
1453 LASSERT(lli->lli_posix_acl == NULL);
1454 if (lli->lli_remote_perms) {
1455 free_rmtperm_hash(lli->lli_remote_perms);
1456 lli->lli_remote_perms = NULL;
1459 #ifdef CONFIG_FS_POSIX_ACL
1460 else if (lli->lli_posix_acl) {
1461 LASSERT(atomic_read(&lli->lli_posix_acl->a_refcount) == 1);
1462 LASSERT(lli->lli_remote_perms == NULL);
1463 posix_acl_release(lli->lli_posix_acl);
1464 lli->lli_posix_acl = NULL;
1467 lli->lli_inode_magic = LLI_INODE_DEAD;
1469 ll_clear_inode_capas(inode);
1470 if (S_ISDIR(inode->i_mode))
1471 ll_dir_clear_lsm_md(inode);
1472 else if (S_ISREG(inode->i_mode) && !is_bad_inode(inode))
1473 LASSERT(list_empty(&lli->lli_agl_list));
1476 * XXX This has to be done before lsm is freed below, because
1477 * cl_object still uses inode lsm.
1479 cl_inode_fini(inode);
1480 lli->lli_has_smd = false;
1485 static int ll_md_setattr(struct dentry *dentry, struct md_op_data *op_data,
1486 struct md_open_data **mod)
1488 struct lustre_md md;
1489 struct inode *inode = dentry->d_inode;
1490 struct ll_sb_info *sbi = ll_i2sbi(inode);
1491 struct ptlrpc_request *request = NULL;
1495 op_data = ll_prep_md_op_data(op_data, inode, NULL, NULL, 0, 0,
1496 LUSTRE_OPC_ANY, NULL);
1497 if (IS_ERR(op_data))
1498 RETURN(PTR_ERR(op_data));
1500 rc = md_setattr(sbi->ll_md_exp, op_data, NULL, 0, NULL, 0,
1503 ptlrpc_req_finished(request);
1504 if (rc == -ENOENT) {
1506 /* Unlinked special device node? Or just a race?
1507 * Pretend we done everything. */
1508 if (!S_ISREG(inode->i_mode) &&
1509 !S_ISDIR(inode->i_mode)) {
1510 ia_valid = op_data->op_attr.ia_valid;
1511 op_data->op_attr.ia_valid &= ~TIMES_SET_FLAGS;
1512 rc = simple_setattr(dentry, &op_data->op_attr);
1513 op_data->op_attr.ia_valid = ia_valid;
1515 } else if (rc != -EPERM && rc != -EACCES && rc != -ETXTBSY) {
1516 CERROR("md_setattr fails: rc = %d\n", rc);
1521 rc = md_get_lustre_md(sbi->ll_md_exp, request, sbi->ll_dt_exp,
1522 sbi->ll_md_exp, &md);
1524 ptlrpc_req_finished(request);
1528 ia_valid = op_data->op_attr.ia_valid;
1529 /* inode size will be in ll_setattr_ost, can't do it now since dirty
1530 * cache is not cleared yet. */
1531 op_data->op_attr.ia_valid &= ~(TIMES_SET_FLAGS | ATTR_SIZE);
1532 rc = simple_setattr(dentry, &op_data->op_attr);
1533 op_data->op_attr.ia_valid = ia_valid;
1535 rc = ll_update_inode(inode, &md);
1536 ptlrpc_req_finished(request);
1541 static int ll_setattr_ost(struct inode *inode, struct iattr *attr)
1543 struct obd_capa *capa;
1546 if (attr->ia_valid & ATTR_SIZE)
1547 capa = ll_osscapa_get(inode, CAPA_OPC_OSS_TRUNC);
1549 capa = ll_mdscapa_get(inode);
1551 rc = cl_setattr_ost(inode, attr, capa);
1553 if (attr->ia_valid & ATTR_SIZE)
1554 ll_truncate_free_capa(capa);
1561 /* If this inode has objects allocated to it (lsm != NULL), then the OST
1562 * object(s) determine the file size and mtime. Otherwise, the MDS will
1563 * keep these values until such a time that objects are allocated for it.
1564 * We do the MDS operations first, as it is checking permissions for us.
1565 * We don't to the MDS RPC if there is nothing that we want to store there,
1566 * otherwise there is no harm in updating mtime/atime on the MDS if we are
1567 * going to do an RPC anyways.
1569 * If we are doing a truncate, we will send the mtime and ctime updates
1570 * to the OST with the punch RPC, otherwise we do an explicit setattr RPC.
1571 * I don't believe it is possible to get e.g. ATTR_MTIME_SET and ATTR_SIZE
1574 * In case of HSMimport, we only set attr on MDS.
1576 int ll_setattr_raw(struct dentry *dentry, struct iattr *attr, bool hsm_import)
1578 struct inode *inode = dentry->d_inode;
1579 struct ll_inode_info *lli = ll_i2info(inode);
1580 struct md_op_data *op_data = NULL;
1581 struct md_open_data *mod = NULL;
1582 bool file_is_released = false;
1586 CDEBUG(D_VFSTRACE, "%s: setattr inode "DFID"(%p) from %llu to %llu, "
1587 "valid %x, hsm_import %d\n",
1588 ll_get_fsname(inode->i_sb, NULL, 0), PFID(&lli->lli_fid),
1589 inode, i_size_read(inode), attr->ia_size, attr->ia_valid,
1592 if (attr->ia_valid & ATTR_SIZE) {
1593 /* Check new size against VFS/VM file size limit and rlimit */
1594 rc = inode_newsize_ok(inode, attr->ia_size);
1598 /* The maximum Lustre file size is variable, based on the
1599 * OST maximum object size and number of stripes. This
1600 * needs another check in addition to the VFS check above. */
1601 if (attr->ia_size > ll_file_maxbytes(inode)) {
1602 CDEBUG(D_INODE,"file "DFID" too large %llu > "LPU64"\n",
1603 PFID(&lli->lli_fid), attr->ia_size,
1604 ll_file_maxbytes(inode));
1608 attr->ia_valid |= ATTR_MTIME | ATTR_CTIME;
1611 /* POSIX: check before ATTR_*TIME_SET set (from inode_change_ok) */
1612 if (attr->ia_valid & TIMES_SET_FLAGS) {
1613 if ((!uid_eq(current_fsuid(), inode->i_uid)) &&
1614 !cfs_capable(CFS_CAP_FOWNER))
1618 /* We mark all of the fields "set" so MDS/OST does not re-set them */
1619 if (attr->ia_valid & ATTR_CTIME) {
1620 attr->ia_ctime = CFS_CURRENT_TIME;
1621 attr->ia_valid |= ATTR_CTIME_SET;
1623 if (!(attr->ia_valid & ATTR_ATIME_SET) &&
1624 (attr->ia_valid & ATTR_ATIME)) {
1625 attr->ia_atime = CFS_CURRENT_TIME;
1626 attr->ia_valid |= ATTR_ATIME_SET;
1628 if (!(attr->ia_valid & ATTR_MTIME_SET) &&
1629 (attr->ia_valid & ATTR_MTIME)) {
1630 attr->ia_mtime = CFS_CURRENT_TIME;
1631 attr->ia_valid |= ATTR_MTIME_SET;
1634 if (attr->ia_valid & (ATTR_MTIME | ATTR_CTIME))
1635 CDEBUG(D_INODE, "setting mtime %lu, ctime %lu, now = %lu\n",
1636 LTIME_S(attr->ia_mtime), LTIME_S(attr->ia_ctime),
1637 cfs_time_current_sec());
1639 /* We always do an MDS RPC, even if we're only changing the size;
1640 * only the MDS knows whether truncate() should fail with -ETXTBUSY */
1642 OBD_ALLOC_PTR(op_data);
1643 if (op_data == NULL)
1646 if (!S_ISDIR(inode->i_mode)) {
1647 if (attr->ia_valid & ATTR_SIZE)
1648 inode_dio_write_done(inode);
1649 mutex_unlock(&inode->i_mutex);
1652 /* truncate on a released file must failed with -ENODATA,
1653 * so size must not be set on MDS for released file
1654 * but other attributes must be set
1656 if (S_ISREG(inode->i_mode)) {
1657 struct lov_stripe_md *lsm;
1660 ll_layout_refresh(inode, &gen);
1661 lsm = ccc_inode_lsm_get(inode);
1662 if (lsm && lsm->lsm_pattern & LOV_PATTERN_F_RELEASED)
1663 file_is_released = true;
1664 ccc_inode_lsm_put(inode, lsm);
1666 if (!hsm_import && attr->ia_valid & ATTR_SIZE) {
1667 if (file_is_released) {
1668 rc = ll_layout_restore(inode, 0, attr->ia_size);
1672 file_is_released = false;
1673 ll_layout_refresh(inode, &gen);
1676 /* If we are changing file size, file content is
1677 * modified, flag it. */
1678 attr->ia_valid |= MDS_OPEN_OWNEROVERRIDE;
1679 spin_lock(&lli->lli_lock);
1680 lli->lli_flags |= LLIF_DATA_MODIFIED;
1681 spin_unlock(&lli->lli_lock);
1682 op_data->op_bias |= MDS_DATA_MODIFIED;
1686 memcpy(&op_data->op_attr, attr, sizeof(*attr));
1688 rc = ll_md_setattr(dentry, op_data, &mod);
1692 /* RPC to MDT is sent, cancel data modification flag */
1693 if (rc == 0 && (op_data->op_bias & MDS_DATA_MODIFIED)) {
1694 spin_lock(&lli->lli_lock);
1695 lli->lli_flags &= ~LLIF_DATA_MODIFIED;
1696 spin_unlock(&lli->lli_lock);
1699 if (!S_ISREG(inode->i_mode) || file_is_released)
1702 if (attr->ia_valid & (ATTR_SIZE |
1703 ATTR_ATIME | ATTR_ATIME_SET |
1704 ATTR_MTIME | ATTR_MTIME_SET)) {
1705 /* For truncate and utimes sending attributes to OSTs, setting
1706 * mtime/atime to the past will be performed under PW [0:EOF]
1707 * extent lock (new_size:EOF for truncate). It may seem
1708 * excessive to send mtime/atime updates to OSTs when not
1709 * setting times to past, but it is necessary due to possible
1710 * time de-synchronization between MDT inode and OST objects */
1711 if (attr->ia_valid & ATTR_SIZE)
1712 down_write(&lli->lli_trunc_sem);
1713 rc = ll_setattr_ost(inode, attr);
1714 if (attr->ia_valid & ATTR_SIZE)
1715 up_write(&lli->lli_trunc_sem);
1719 if (op_data != NULL)
1720 ll_finish_md_op_data(op_data);
1722 if (!S_ISDIR(inode->i_mode)) {
1723 mutex_lock(&inode->i_mutex);
1724 if ((attr->ia_valid & ATTR_SIZE) && !hsm_import)
1725 inode_dio_wait(inode);
1728 ll_stats_ops_tally(ll_i2sbi(inode), (attr->ia_valid & ATTR_SIZE) ?
1729 LPROC_LL_TRUNC : LPROC_LL_SETATTR, 1);
1734 int ll_setattr(struct dentry *de, struct iattr *attr)
1736 int mode = de->d_inode->i_mode;
1738 if ((attr->ia_valid & (ATTR_CTIME|ATTR_SIZE|ATTR_MODE)) ==
1739 (ATTR_CTIME|ATTR_SIZE|ATTR_MODE))
1740 attr->ia_valid |= MDS_OPEN_OWNEROVERRIDE;
1742 if (((attr->ia_valid & (ATTR_MODE|ATTR_FORCE|ATTR_SIZE)) ==
1743 (ATTR_SIZE|ATTR_MODE)) &&
1744 (((mode & S_ISUID) && !(attr->ia_mode & S_ISUID)) ||
1745 (((mode & (S_ISGID|S_IXGRP)) == (S_ISGID|S_IXGRP)) &&
1746 !(attr->ia_mode & S_ISGID))))
1747 attr->ia_valid |= ATTR_FORCE;
1749 if ((attr->ia_valid & ATTR_MODE) &&
1751 !(attr->ia_mode & S_ISUID) &&
1752 !(attr->ia_valid & ATTR_KILL_SUID))
1753 attr->ia_valid |= ATTR_KILL_SUID;
1755 if ((attr->ia_valid & ATTR_MODE) &&
1756 ((mode & (S_ISGID|S_IXGRP)) == (S_ISGID|S_IXGRP)) &&
1757 !(attr->ia_mode & S_ISGID) &&
1758 !(attr->ia_valid & ATTR_KILL_SGID))
1759 attr->ia_valid |= ATTR_KILL_SGID;
1761 return ll_setattr_raw(de, attr, false);
1764 int ll_statfs_internal(struct super_block *sb, struct obd_statfs *osfs,
1765 __u64 max_age, __u32 flags)
1767 struct ll_sb_info *sbi = ll_s2sbi(sb);
1768 struct obd_statfs obd_osfs;
1772 rc = obd_statfs(NULL, sbi->ll_md_exp, osfs, max_age, flags);
1774 CERROR("md_statfs fails: rc = %d\n", rc);
1778 osfs->os_type = sb->s_magic;
1780 CDEBUG(D_SUPER, "MDC blocks "LPU64"/"LPU64" objects "LPU64"/"LPU64"\n",
1781 osfs->os_bavail, osfs->os_blocks, osfs->os_ffree,osfs->os_files);
1783 if (sbi->ll_flags & LL_SBI_LAZYSTATFS)
1784 flags |= OBD_STATFS_NODELAY;
1786 rc = obd_statfs_rqset(sbi->ll_dt_exp, &obd_osfs, max_age, flags);
1788 CERROR("obd_statfs fails: rc = %d\n", rc);
1792 CDEBUG(D_SUPER, "OSC blocks "LPU64"/"LPU64" objects "LPU64"/"LPU64"\n",
1793 obd_osfs.os_bavail, obd_osfs.os_blocks, obd_osfs.os_ffree,
1796 osfs->os_bsize = obd_osfs.os_bsize;
1797 osfs->os_blocks = obd_osfs.os_blocks;
1798 osfs->os_bfree = obd_osfs.os_bfree;
1799 osfs->os_bavail = obd_osfs.os_bavail;
1801 /* If we don't have as many objects free on the OST as inodes
1802 * on the MDS, we reduce the total number of inodes to
1803 * compensate, so that the "inodes in use" number is correct.
1805 if (obd_osfs.os_ffree < osfs->os_ffree) {
1806 osfs->os_files = (osfs->os_files - osfs->os_ffree) +
1808 osfs->os_ffree = obd_osfs.os_ffree;
1813 int ll_statfs(struct dentry *de, struct kstatfs *sfs)
1815 struct super_block *sb = de->d_sb;
1816 struct obd_statfs osfs;
1817 __u64 fsid = huge_encode_dev(sb->s_dev);
1820 CDEBUG(D_VFSTRACE, "VFS Op: at "LPU64" jiffies\n", get_jiffies_64());
1821 ll_stats_ops_tally(ll_s2sbi(sb), LPROC_LL_STAFS, 1);
1823 /* Some amount of caching on the client is allowed */
1824 rc = ll_statfs_internal(sb, &osfs,
1825 cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS),
1830 statfs_unpack(sfs, &osfs);
1832 /* We need to downshift for all 32-bit kernels, because we can't
1833 * tell if the kernel is being called via sys_statfs64() or not.
1834 * Stop before overflowing f_bsize - in which case it is better
1835 * to just risk EOVERFLOW if caller is using old sys_statfs(). */
1836 if (sizeof(long) < 8) {
1837 while (osfs.os_blocks > ~0UL && sfs->f_bsize < 0x40000000) {
1840 osfs.os_blocks >>= 1;
1841 osfs.os_bfree >>= 1;
1842 osfs.os_bavail >>= 1;
1846 sfs->f_blocks = osfs.os_blocks;
1847 sfs->f_bfree = osfs.os_bfree;
1848 sfs->f_bavail = osfs.os_bavail;
1849 sfs->f_fsid.val[0] = (__u32)fsid;
1850 sfs->f_fsid.val[1] = (__u32)(fsid >> 32);
1854 void ll_inode_size_lock(struct inode *inode)
1856 struct ll_inode_info *lli;
1858 LASSERT(!S_ISDIR(inode->i_mode));
1860 lli = ll_i2info(inode);
1861 mutex_lock(&lli->lli_size_mutex);
1864 void ll_inode_size_unlock(struct inode *inode)
1866 struct ll_inode_info *lli;
1868 lli = ll_i2info(inode);
1869 mutex_unlock(&lli->lli_size_mutex);
1872 int ll_update_inode(struct inode *inode, struct lustre_md *md)
1874 struct ll_inode_info *lli = ll_i2info(inode);
1875 struct mdt_body *body = md->body;
1876 struct lov_stripe_md *lsm = md->lsm;
1877 struct ll_sb_info *sbi = ll_i2sbi(inode);
1879 LASSERT((lsm != NULL) == ((body->mbo_valid & OBD_MD_FLEASIZE) != 0));
1881 if (!lli->lli_has_smd &&
1882 !(sbi->ll_flags & LL_SBI_LAYOUT_LOCK))
1883 cl_file_inode_init(inode, md);
1885 lli->lli_maxbytes = lsm->lsm_maxbytes;
1886 if (lli->lli_maxbytes > MAX_LFS_FILESIZE)
1887 lli->lli_maxbytes = MAX_LFS_FILESIZE;
1890 if (S_ISDIR(inode->i_mode)) {
1893 rc = ll_update_lsm_md(inode, md);
1898 if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
1899 if (body->mbo_valid & OBD_MD_FLRMTPERM)
1900 ll_update_remote_perm(inode, md->remote_perm);
1902 #ifdef CONFIG_FS_POSIX_ACL
1903 else if (body->mbo_valid & OBD_MD_FLACL) {
1904 spin_lock(&lli->lli_lock);
1905 if (lli->lli_posix_acl)
1906 posix_acl_release(lli->lli_posix_acl);
1907 lli->lli_posix_acl = md->posix_acl;
1908 spin_unlock(&lli->lli_lock);
1911 inode->i_ino = cl_fid_build_ino(&body->mbo_fid1,
1912 sbi->ll_flags & LL_SBI_32BIT_API);
1913 inode->i_generation = cl_fid_build_gen(&body->mbo_fid1);
1915 if (body->mbo_valid & OBD_MD_FLATIME) {
1916 if (body->mbo_atime > LTIME_S(inode->i_atime))
1917 LTIME_S(inode->i_atime) = body->mbo_atime;
1918 lli->lli_atime = body->mbo_atime;
1921 if (body->mbo_valid & OBD_MD_FLMTIME) {
1922 if (body->mbo_mtime > LTIME_S(inode->i_mtime)) {
1923 CDEBUG(D_INODE, "setting ino %lu mtime from %lu "
1924 "to "LPU64"\n", inode->i_ino,
1925 LTIME_S(inode->i_mtime), body->mbo_mtime);
1926 LTIME_S(inode->i_mtime) = body->mbo_mtime;
1928 lli->lli_mtime = body->mbo_mtime;
1931 if (body->mbo_valid & OBD_MD_FLCTIME) {
1932 if (body->mbo_ctime > LTIME_S(inode->i_ctime))
1933 LTIME_S(inode->i_ctime) = body->mbo_ctime;
1934 lli->lli_ctime = body->mbo_ctime;
1937 if (body->mbo_valid & OBD_MD_FLMODE)
1938 inode->i_mode = (inode->i_mode & S_IFMT) |
1939 (body->mbo_mode & ~S_IFMT);
1941 if (body->mbo_valid & OBD_MD_FLTYPE)
1942 inode->i_mode = (inode->i_mode & ~S_IFMT) |
1943 (body->mbo_mode & S_IFMT);
1945 LASSERT(inode->i_mode != 0);
1946 if (S_ISREG(inode->i_mode))
1947 inode->i_blkbits = min(PTLRPC_MAX_BRW_BITS + 1,
1948 LL_MAX_BLKSIZE_BITS);
1950 inode->i_blkbits = inode->i_sb->s_blocksize_bits;
1952 if (body->mbo_valid & OBD_MD_FLUID)
1953 inode->i_uid = make_kuid(&init_user_ns, body->mbo_uid);
1954 if (body->mbo_valid & OBD_MD_FLGID)
1955 inode->i_gid = make_kgid(&init_user_ns, body->mbo_gid);
1956 if (body->mbo_valid & OBD_MD_FLFLAGS)
1957 inode->i_flags = ll_ext_to_inode_flags(body->mbo_flags);
1958 if (body->mbo_valid & OBD_MD_FLNLINK)
1959 set_nlink(inode, body->mbo_nlink);
1960 if (body->mbo_valid & OBD_MD_FLRDEV)
1961 inode->i_rdev = old_decode_dev(body->mbo_rdev);
1963 if (body->mbo_valid & OBD_MD_FLID) {
1964 /* FID shouldn't be changed! */
1965 if (fid_is_sane(&lli->lli_fid)) {
1966 LASSERTF(lu_fid_eq(&lli->lli_fid, &body->mbo_fid1),
1967 "Trying to change FID "DFID
1968 " to the "DFID", inode "DFID"(%p)\n",
1969 PFID(&lli->lli_fid), PFID(&body->mbo_fid1),
1970 PFID(ll_inode2fid(inode)), inode);
1972 lli->lli_fid = body->mbo_fid1;
1976 LASSERT(fid_seq(&lli->lli_fid) != 0);
1978 if (body->mbo_valid & OBD_MD_FLSIZE) {
1979 i_size_write(inode, body->mbo_size);
1981 CDEBUG(D_VFSTRACE, "inode="DFID", updating i_size %llu\n",
1982 PFID(ll_inode2fid(inode)),
1983 (unsigned long long)body->mbo_size);
1985 if (body->mbo_valid & OBD_MD_FLBLOCKS)
1986 inode->i_blocks = body->mbo_blocks;
1989 if (body->mbo_valid & OBD_MD_FLMDSCAPA) {
1990 LASSERT(md->mds_capa);
1991 ll_add_capa(inode, md->mds_capa);
1994 if (body->mbo_valid & OBD_MD_FLOSSCAPA) {
1995 LASSERT(md->oss_capa);
1996 ll_add_capa(inode, md->oss_capa);
1999 if (body->mbo_valid & OBD_MD_TSTATE) {
2000 if (body->mbo_t_state & MS_RESTORE)
2001 lli->lli_flags |= LLIF_FILE_RESTORING;
2007 int ll_read_inode2(struct inode *inode, void *opaque)
2009 struct lustre_md *md = opaque;
2010 struct ll_inode_info *lli = ll_i2info(inode);
2014 CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p)\n",
2015 PFID(&lli->lli_fid), inode);
2017 LASSERT(!lli->lli_has_smd);
2019 /* Core attributes from the MDS first. This is a new inode, and
2020 * the VFS doesn't zero times in the core inode so we have to do
2021 * it ourselves. They will be overwritten by either MDS or OST
2022 * attributes - we just need to make sure they aren't newer. */
2023 LTIME_S(inode->i_mtime) = 0;
2024 LTIME_S(inode->i_atime) = 0;
2025 LTIME_S(inode->i_ctime) = 0;
2027 rc = ll_update_inode(inode, md);
2031 /* OIDEBUG(inode); */
2033 /* initializing backing dev info. */
2034 inode->i_mapping->backing_dev_info = &s2lsi(inode->i_sb)->lsi_bdi;
2037 if (S_ISREG(inode->i_mode)) {
2038 struct ll_sb_info *sbi = ll_i2sbi(inode);
2039 inode->i_op = &ll_file_inode_operations;
2040 inode->i_fop = sbi->ll_fop;
2041 inode->i_mapping->a_ops = (struct address_space_operations *)&ll_aops;
2043 } else if (S_ISDIR(inode->i_mode)) {
2044 inode->i_op = &ll_dir_inode_operations;
2045 inode->i_fop = &ll_dir_operations;
2047 } else if (S_ISLNK(inode->i_mode)) {
2048 inode->i_op = &ll_fast_symlink_inode_operations;
2051 inode->i_op = &ll_special_inode_operations;
2053 init_special_inode(inode, inode->i_mode,
2062 void ll_delete_inode(struct inode *inode)
2064 struct ll_inode_info *lli = ll_i2info(inode);
2067 if (S_ISREG(inode->i_mode) && lli->lli_clob != NULL)
2068 /* It is last chance to write out dirty pages,
2069 * otherwise we may lose data while umount */
2070 cl_sync_file_range(inode, 0, OBD_OBJECT_EOF, CL_FSYNC_LOCAL, 1);
2072 truncate_inode_pages_final(&inode->i_data);
2074 LASSERTF(inode->i_data.nrpages == 0, "inode="DFID"(%p) nrpages=%lu, "
2075 "see https://jira.hpdd.intel.com/browse/LU-118\n",
2076 PFID(ll_inode2fid(inode)), inode, inode->i_data.nrpages);
2078 #ifdef HAVE_SBOPS_EVICT_INODE
2079 ll_clear_inode(inode);
2086 int ll_iocontrol(struct inode *inode, struct file *file,
2087 unsigned int cmd, unsigned long arg)
2089 struct ll_sb_info *sbi = ll_i2sbi(inode);
2090 struct ptlrpc_request *req = NULL;
2095 case FSFILT_IOC_GETFLAGS: {
2096 struct mdt_body *body;
2097 struct md_op_data *op_data;
2099 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL,
2100 0, 0, LUSTRE_OPC_ANY,
2102 if (IS_ERR(op_data))
2103 RETURN(PTR_ERR(op_data));
2105 op_data->op_valid = OBD_MD_FLFLAGS;
2106 rc = md_getattr(sbi->ll_md_exp, op_data, &req);
2107 ll_finish_md_op_data(op_data);
2109 CERROR("%s: failure inode "DFID": rc = %d\n",
2110 sbi->ll_md_exp->exp_obd->obd_name,
2111 PFID(ll_inode2fid(inode)), rc);
2115 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
2117 flags = body->mbo_flags;
2119 ptlrpc_req_finished(req);
2121 RETURN(put_user(flags, (int __user *)arg));
2123 case FSFILT_IOC_SETFLAGS: {
2124 struct lov_stripe_md *lsm;
2125 struct obd_info oinfo = { { { 0 } } };
2126 struct md_op_data *op_data;
2128 if (get_user(flags, (int __user *)arg))
2131 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
2132 LUSTRE_OPC_ANY, NULL);
2133 if (IS_ERR(op_data))
2134 RETURN(PTR_ERR(op_data));
2136 op_data->op_attr_flags = flags;
2137 op_data->op_attr.ia_valid |= ATTR_ATTR_FLAG;
2138 rc = md_setattr(sbi->ll_md_exp, op_data,
2139 NULL, 0, NULL, 0, &req, NULL);
2140 ll_finish_md_op_data(op_data);
2141 ptlrpc_req_finished(req);
2145 inode->i_flags = ll_ext_to_inode_flags(flags);
2147 lsm = ccc_inode_lsm_get(inode);
2148 if (!lsm_has_objects(lsm)) {
2149 ccc_inode_lsm_put(inode, lsm);
2153 OBDO_ALLOC(oinfo.oi_oa);
2155 ccc_inode_lsm_put(inode, lsm);
2159 oinfo.oi_oa->o_oi = lsm->lsm_oi;
2160 oinfo.oi_oa->o_flags = flags;
2161 oinfo.oi_oa->o_valid = OBD_MD_FLID | OBD_MD_FLFLAGS |
2163 oinfo.oi_capa = ll_mdscapa_get(inode);
2164 obdo_set_parent_fid(oinfo.oi_oa, &ll_i2info(inode)->lli_fid);
2165 rc = obd_setattr_rqset(sbi->ll_dt_exp, &oinfo, NULL);
2166 capa_put(oinfo.oi_capa);
2167 OBDO_FREE(oinfo.oi_oa);
2168 ccc_inode_lsm_put(inode, lsm);
2170 if (rc && rc != -EPERM && rc != -EACCES)
2171 CERROR("osc_setattr_async fails: rc = %d\n", rc);
2182 int ll_flush_ctx(struct inode *inode)
2184 struct ll_sb_info *sbi = ll_i2sbi(inode);
2186 CDEBUG(D_SEC, "flush context for user %d\n",
2187 from_kuid(&init_user_ns, current_uid()));
2189 obd_set_info_async(NULL, sbi->ll_md_exp,
2190 sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
2192 obd_set_info_async(NULL, sbi->ll_dt_exp,
2193 sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
2198 /* umount -f client means force down, don't save state */
2199 void ll_umount_begin(struct super_block *sb)
2201 struct ll_sb_info *sbi = ll_s2sbi(sb);
2202 struct obd_device *obd;
2203 struct obd_ioctl_data *ioc_data;
2206 CDEBUG(D_VFSTRACE, "VFS Op: superblock %p count %d active %d\n", sb,
2207 sb->s_count, atomic_read(&sb->s_active));
2209 obd = class_exp2obd(sbi->ll_md_exp);
2211 CERROR("Invalid MDC connection handle "LPX64"\n",
2212 sbi->ll_md_exp->exp_handle.h_cookie);
2218 obd = class_exp2obd(sbi->ll_dt_exp);
2220 CERROR("Invalid LOV connection handle "LPX64"\n",
2221 sbi->ll_dt_exp->exp_handle.h_cookie);
2227 OBD_ALLOC_PTR(ioc_data);
2229 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_md_exp,
2230 sizeof *ioc_data, ioc_data, NULL);
2232 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_dt_exp,
2233 sizeof *ioc_data, ioc_data, NULL);
2235 OBD_FREE_PTR(ioc_data);
2238 /* Really, we'd like to wait until there are no requests outstanding,
2239 * and then continue. For now, we just invalidate the requests,
2240 * schedule() and sleep one second if needed, and hope.
2246 int ll_remount_fs(struct super_block *sb, int *flags, char *data)
2248 struct ll_sb_info *sbi = ll_s2sbi(sb);
2249 char *profilenm = get_profile_name(sb);
2253 if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY)) {
2254 read_only = *flags & MS_RDONLY;
2255 err = obd_set_info_async(NULL, sbi->ll_md_exp,
2256 sizeof(KEY_READ_ONLY),
2257 KEY_READ_ONLY, sizeof(read_only),
2260 LCONSOLE_WARN("Failed to remount %s %s (%d)\n",
2261 profilenm, read_only ?
2262 "read-only" : "read-write", err);
2267 sb->s_flags |= MS_RDONLY;
2269 sb->s_flags &= ~MS_RDONLY;
2271 if (sbi->ll_flags & LL_SBI_VERBOSE)
2272 LCONSOLE_WARN("Remounted %s %s\n", profilenm,
2273 read_only ? "read-only" : "read-write");
2279 * Cleanup the open handle that is cached on MDT-side.
2281 * For open case, the client side open handling thread may hit error
2282 * after the MDT grant the open. Under such case, the client should
2283 * send close RPC to the MDT as cleanup; otherwise, the open handle
2284 * on the MDT will be leaked there until the client umount or evicted.
2286 * In further, if someone unlinked the file, because the open handle
2287 * holds the reference on such file/object, then it will block the
2288 * subsequent threads that want to locate such object via FID.
2290 * \param[in] sb super block for this file-system
2291 * \param[in] open_req pointer to the original open request
2293 void ll_open_cleanup(struct super_block *sb, struct ptlrpc_request *open_req)
2295 struct mdt_body *body;
2296 struct md_op_data *op_data;
2297 struct ptlrpc_request *close_req = NULL;
2298 struct obd_export *exp = ll_s2sbi(sb)->ll_md_exp;
2301 body = req_capsule_server_get(&open_req->rq_pill, &RMF_MDT_BODY);
2302 OBD_ALLOC_PTR(op_data);
2303 if (op_data == NULL) {
2304 CWARN("%s: cannot allocate op_data to release open handle for "
2306 ll_get_fsname(sb, NULL, 0), PFID(&body->mbo_fid1));
2311 op_data->op_fid1 = body->mbo_fid1;
2312 op_data->op_handle = body->mbo_handle;
2313 op_data->op_mod_time = cfs_time_current_sec();
2314 md_close(exp, op_data, NULL, &close_req);
2315 ptlrpc_req_finished(close_req);
2316 ll_finish_md_op_data(op_data);
2321 int ll_prep_inode(struct inode **inode, struct ptlrpc_request *req,
2322 struct super_block *sb, struct lookup_intent *it)
2324 struct ll_sb_info *sbi = NULL;
2325 struct lustre_md md = { NULL };
2329 LASSERT(*inode || sb);
2330 sbi = sb ? ll_s2sbi(sb) : ll_i2sbi(*inode);
2331 rc = md_get_lustre_md(sbi->ll_md_exp, req, sbi->ll_dt_exp,
2332 sbi->ll_md_exp, &md);
2337 rc = ll_update_inode(*inode, &md);
2341 LASSERT(sb != NULL);
2344 * At this point server returns to client's same fid as client
2345 * generated for creating. So using ->fid1 is okay here.
2347 LASSERT(fid_is_sane(&md.body->mbo_fid1));
2349 *inode = ll_iget(sb, cl_fid_build_ino(&md.body->mbo_fid1,
2350 sbi->ll_flags & LL_SBI_32BIT_API),
2352 if (IS_ERR(*inode)) {
2353 #ifdef CONFIG_FS_POSIX_ACL
2355 posix_acl_release(md.posix_acl);
2356 md.posix_acl = NULL;
2359 rc = IS_ERR(*inode) ? PTR_ERR(*inode) : -ENOMEM;
2361 CERROR("new_inode -fatal: rc %d\n", rc);
2366 /* Handling piggyback layout lock.
2367 * Layout lock can be piggybacked by getattr and open request.
2368 * The lsm can be applied to inode only if it comes with a layout lock
2369 * otherwise correct layout may be overwritten, for example:
2370 * 1. proc1: mdt returns a lsm but not granting layout
2371 * 2. layout was changed by another client
2372 * 3. proc2: refresh layout and layout lock granted
2373 * 4. proc1: to apply a stale layout */
2374 if (it != NULL && it->d.lustre.it_lock_mode != 0) {
2375 struct lustre_handle lockh;
2376 struct ldlm_lock *lock;
2378 lockh.cookie = it->d.lustre.it_lock_handle;
2379 lock = ldlm_handle2lock(&lockh);
2380 LASSERT(lock != NULL);
2381 if (ldlm_has_layout(lock)) {
2382 struct cl_object_conf conf;
2384 memset(&conf, 0, sizeof(conf));
2385 conf.coc_opc = OBJECT_CONF_SET;
2386 conf.coc_inode = *inode;
2387 conf.coc_lock = lock;
2388 conf.u.coc_md = &md;
2389 (void)ll_layout_conf(*inode, &conf);
2391 LDLM_LOCK_PUT(lock);
2398 obd_free_memmd(sbi->ll_dt_exp, &md.lsm);
2399 md_free_lustre_md(sbi->ll_md_exp, &md);
2402 if (rc != 0 && it != NULL && it->it_op & IT_OPEN)
2403 ll_open_cleanup(sb != NULL ? sb : (*inode)->i_sb, req);
2408 int ll_obd_statfs(struct inode *inode, void __user *arg)
2410 struct ll_sb_info *sbi = NULL;
2411 struct obd_export *exp;
2413 struct obd_ioctl_data *data = NULL;
2415 __u32 __user flags; /* not user, but obd_iocontrol is abused */
2418 if (!inode || !(sbi = ll_i2sbi(inode)))
2419 GOTO(out_statfs, rc = -EINVAL);
2421 rc = obd_ioctl_getdata(&buf, &len, arg);
2423 GOTO(out_statfs, rc);
2426 if (!data->ioc_inlbuf1 || !data->ioc_inlbuf2 ||
2427 !data->ioc_pbuf1 || !data->ioc_pbuf2)
2428 GOTO(out_statfs, rc = -EINVAL);
2430 if (data->ioc_inllen1 != sizeof(__u32) ||
2431 data->ioc_inllen2 != sizeof(__u32) ||
2432 data->ioc_plen1 != sizeof(struct obd_statfs) ||
2433 data->ioc_plen2 != sizeof(struct obd_uuid))
2434 GOTO(out_statfs, rc = -EINVAL);
2436 memcpy(&type, data->ioc_inlbuf1, sizeof(__u32));
2437 if (type & LL_STATFS_LMV)
2438 exp = sbi->ll_md_exp;
2439 else if (type & LL_STATFS_LOV)
2440 exp = sbi->ll_dt_exp;
2442 GOTO(out_statfs, rc = -ENODEV);
2444 flags = (type & LL_STATFS_NODELAY) ? OBD_STATFS_NODELAY : 0;
2445 rc = obd_iocontrol(IOC_OBD_STATFS, exp, len, buf, &flags);
2447 GOTO(out_statfs, rc);
2450 obd_ioctl_freedata(buf, len);
2454 int ll_process_config(struct lustre_cfg *lcfg)
2456 struct super_block *sb;
2461 /* The instance name contains the sb: lustre-client-aacfe000 */
2462 ptr = strrchr(lustre_cfg_string(lcfg, 0), '-');
2463 if (!ptr || !*(++ptr))
2465 if (sscanf(ptr, "%lx", &x) != 1)
2467 sb = (struct super_block *)x;
2468 /* This better be a real Lustre superblock! */
2469 LASSERT(s2lsi(sb)->lsi_lmd->lmd_magic == LMD_MAGIC);
2471 /* Note we have not called client_common_fill_super yet, so
2472 proc fns must be able to handle that! */
2473 rc = class_process_proc_param(PARAM_LLITE, lprocfs_llite_obd_vars,
2480 /* this function prepares md_op_data hint for passing ot down to MD stack. */
2481 struct md_op_data * ll_prep_md_op_data(struct md_op_data *op_data,
2482 struct inode *i1, struct inode *i2,
2483 const char *name, size_t namelen,
2484 __u32 mode, __u32 opc, void *data)
2486 LASSERT(i1 != NULL);
2489 /* Do not reuse namelen for something else. */
2491 return ERR_PTR(-EINVAL);
2493 if (namelen > ll_i2sbi(i1)->ll_namelen)
2494 return ERR_PTR(-ENAMETOOLONG);
2496 if (!lu_name_is_valid_2(name, namelen))
2497 return ERR_PTR(-EINVAL);
2500 if (op_data == NULL)
2501 OBD_ALLOC_PTR(op_data);
2503 if (op_data == NULL)
2504 return ERR_PTR(-ENOMEM);
2506 ll_i2gids(op_data->op_suppgids, i1, i2);
2507 op_data->op_fid1 = *ll_inode2fid(i1);
2508 op_data->op_capa1 = ll_mdscapa_get(i1);
2509 op_data->op_default_stripe_offset = -1;
2510 if (S_ISDIR(i1->i_mode)) {
2511 op_data->op_mea1 = ll_i2info(i1)->lli_lsm_md;
2512 op_data->op_default_stripe_offset =
2513 ll_i2info(i1)->lli_def_stripe_offset;
2517 op_data->op_fid2 = *ll_inode2fid(i2);
2518 op_data->op_capa2 = ll_mdscapa_get(i2);
2519 if (S_ISDIR(i2->i_mode))
2520 op_data->op_mea2 = ll_i2info(i2)->lli_lsm_md;
2522 fid_zero(&op_data->op_fid2);
2523 op_data->op_capa2 = NULL;
2526 if (ll_i2sbi(i1)->ll_flags & LL_SBI_64BIT_HASH)
2527 op_data->op_cli_flags |= CLI_HASH64;
2529 if (ll_need_32bit_api(ll_i2sbi(i1)))
2530 op_data->op_cli_flags |= CLI_API32;
2532 op_data->op_name = name;
2533 op_data->op_namelen = namelen;
2534 op_data->op_mode = mode;
2535 op_data->op_mod_time = cfs_time_current_sec();
2536 op_data->op_fsuid = from_kuid(&init_user_ns, current_fsuid());
2537 op_data->op_fsgid = from_kgid(&init_user_ns, current_fsgid());
2538 op_data->op_cap = cfs_curproc_cap_pack();
2539 op_data->op_bias = 0;
2540 op_data->op_cli_flags = 0;
2541 if ((opc == LUSTRE_OPC_CREATE) && (name != NULL) &&
2542 filename_is_volatile(name, namelen, &op_data->op_mds)) {
2543 op_data->op_bias |= MDS_CREATE_VOLATILE;
2545 op_data->op_mds = 0;
2547 op_data->op_data = data;
2549 /* When called by ll_setattr_raw, file is i1. */
2550 if (LLIF_DATA_MODIFIED & ll_i2info(i1)->lli_flags)
2551 op_data->op_bias |= MDS_DATA_MODIFIED;
2556 void ll_finish_md_op_data(struct md_op_data *op_data)
2558 capa_put(op_data->op_capa1);
2559 capa_put(op_data->op_capa2);
2560 OBD_FREE_PTR(op_data);
2563 #ifdef HAVE_SUPEROPS_USE_DENTRY
2564 int ll_show_options(struct seq_file *seq, struct dentry *dentry)
2566 int ll_show_options(struct seq_file *seq, struct vfsmount *vfs)
2569 struct ll_sb_info *sbi;
2571 #ifdef HAVE_SUPEROPS_USE_DENTRY
2572 LASSERT((seq != NULL) && (dentry != NULL));
2573 sbi = ll_s2sbi(dentry->d_sb);
2575 LASSERT((seq != NULL) && (vfs != NULL));
2576 sbi = ll_s2sbi(vfs->mnt_sb);
2579 if (sbi->ll_flags & LL_SBI_NOLCK)
2580 seq_puts(seq, ",nolock");
2582 if (sbi->ll_flags & LL_SBI_FLOCK)
2583 seq_puts(seq, ",flock");
2585 if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
2586 seq_puts(seq, ",localflock");
2588 if (sbi->ll_flags & LL_SBI_USER_XATTR)
2589 seq_puts(seq, ",user_xattr");
2591 if (sbi->ll_flags & LL_SBI_LAZYSTATFS)
2592 seq_puts(seq, ",lazystatfs");
2594 if (sbi->ll_flags & LL_SBI_USER_FID2PATH)
2595 seq_puts(seq, ",user_fid2path");
2601 * Get obd name by cmd, and copy out to user space
2603 int ll_get_obd_name(struct inode *inode, unsigned int cmd, unsigned long arg)
2605 struct ll_sb_info *sbi = ll_i2sbi(inode);
2606 struct obd_device *obd;
2609 if (cmd == OBD_IOC_GETDTNAME)
2610 obd = class_exp2obd(sbi->ll_dt_exp);
2611 else if (cmd == OBD_IOC_GETMDNAME)
2612 obd = class_exp2obd(sbi->ll_md_exp);
2619 if (copy_to_user((void __user *)arg, obd->obd_name,
2620 strlen(obd->obd_name) + 1))
2627 * Get lustre file system name by \a sbi. If \a buf is provided(non-NULL), the
2628 * fsname will be returned in this buffer; otherwise, a static buffer will be
2629 * used to store the fsname and returned to caller.
2631 char *ll_get_fsname(struct super_block *sb, char *buf, int buflen)
2633 static char fsname_static[MTI_NAME_MAXLEN];
2634 struct lustre_sb_info *lsi = s2lsi(sb);
2639 /* this means the caller wants to use static buffer
2640 * and it doesn't care about race. Usually this is
2641 * in error reporting path */
2642 buf = fsname_static;
2643 buflen = sizeof(fsname_static);
2646 len = strlen(lsi->lsi_lmd->lmd_profile);
2647 ptr = strrchr(lsi->lsi_lmd->lmd_profile, '-');
2648 if (ptr && (strcmp(ptr, "-client") == 0))
2651 if (unlikely(len >= buflen))
2653 strncpy(buf, lsi->lsi_lmd->lmd_profile, len);
2659 static char* ll_d_path(struct dentry *dentry, char *buf, int bufsize)
2666 p.mnt = current->fs->root.mnt;
2668 path = d_path(&p, buf, bufsize);
2673 void ll_dirty_page_discard_warn(struct page *page, int ioret)
2675 char *buf, *path = NULL;
2676 struct dentry *dentry = NULL;
2677 struct vvp_object *obj = cl_inode2vvp(page->mapping->host);
2679 /* this can be called inside spin lock so use GFP_ATOMIC. */
2680 buf = (char *)__get_free_page(GFP_ATOMIC);
2682 dentry = d_find_alias(page->mapping->host);
2684 path = ll_d_path(dentry, buf, PAGE_SIZE);
2688 "%s: dirty page discard: %s/fid: "DFID"/%s may get corrupted "
2689 "(rc %d)\n", ll_get_fsname(page->mapping->host->i_sb, NULL, 0),
2690 s2lsi(page->mapping->host->i_sb)->lsi_lmd->lmd_dev,
2691 PFID(&obj->vob_header.coh_lu.loh_fid),
2692 (path && !IS_ERR(path)) ? path : "", ioret);
2698 free_page((unsigned long)buf);
2701 ssize_t ll_copy_user_md(const struct lov_user_md __user *md,
2702 struct lov_user_md **kbuf)
2704 struct lov_user_md lum;
2708 if (copy_from_user(&lum, md, sizeof(lum)))
2711 lum_size = ll_lov_user_md_size(&lum);
2715 OBD_ALLOC(*kbuf, lum_size);
2719 if (copy_from_user(*kbuf, md, lum_size) != 0) {
2720 OBD_FREE(*kbuf, lum_size);
2728 * Compute llite root squash state after a change of root squash
2729 * configuration setting or add/remove of a lnet nid
2731 void ll_compute_rootsquash_state(struct ll_sb_info *sbi)
2733 struct root_squash_info *squash = &sbi->ll_squash;
2736 lnet_process_id_t id;
2738 /* Update norootsquash flag */
2739 down_write(&squash->rsi_sem);
2740 if (list_empty(&squash->rsi_nosquash_nids))
2741 sbi->ll_flags &= ~LL_SBI_NOROOTSQUASH;
2743 /* Do not apply root squash as soon as one of our NIDs is
2744 * in the nosquash_nids list */
2747 while (LNetGetId(i++, &id) != -ENOENT) {
2748 if (LNET_NETTYP(LNET_NIDNET(id.nid)) == LOLND)
2750 if (cfs_match_nid(id.nid, &squash->rsi_nosquash_nids)) {
2756 sbi->ll_flags |= LL_SBI_NOROOTSQUASH;
2758 sbi->ll_flags &= ~LL_SBI_NOROOTSQUASH;
2760 up_write(&squash->rsi_sem);
2764 * Parse linkea content to extract information about a given hardlink
2766 * \param[in] ldata - Initialized linkea data
2767 * \param[in] linkno - Link identifier
2768 * \param[out] parent_fid - The entry's parent FID
2769 * \param[out] ln - Entry name destination buffer
2771 * \retval 0 on success
2772 * \retval Appropriate negative error code on failure
2774 static int ll_linkea_decode(struct linkea_data *ldata, unsigned int linkno,
2775 struct lu_fid *parent_fid, struct lu_name *ln)
2781 rc = linkea_init(ldata);
2785 if (linkno >= ldata->ld_leh->leh_reccount)
2786 /* beyond last link */
2789 linkea_first_entry(ldata);
2790 for (idx = 0; ldata->ld_lee != NULL; idx++) {
2791 linkea_entry_unpack(ldata->ld_lee, &ldata->ld_reclen, ln,
2796 linkea_next_entry(ldata);
2806 * Get parent FID and name of an identified link. Operation is performed for
2807 * a given link number, letting the caller iterate over linkno to list one or
2808 * all links of an entry.
2810 * \param[in] file - File descriptor against which to perform the operation
2811 * \param[in,out] arg - User-filled structure containing the linkno to operate
2812 * on and the available size. It is eventually filled with
2813 * the requested information or left untouched on error
2815 * \retval - 0 on success
2816 * \retval - Appropriate negative error code on failure
2818 int ll_getparent(struct file *file, struct getparent __user *arg)
2820 struct dentry *dentry = file->f_dentry;
2821 struct inode *inode = file->f_dentry->d_inode;
2822 struct linkea_data *ldata;
2823 struct lu_buf buf = LU_BUF_NULL;
2825 struct lu_fid parent_fid;
2832 if (!cfs_capable(CFS_CAP_DAC_READ_SEARCH) &&
2833 !(ll_i2sbi(inode)->ll_flags & LL_SBI_USER_FID2PATH))
2836 if (get_user(name_size, &arg->gp_name_size))
2839 if (get_user(linkno, &arg->gp_linkno))
2842 if (name_size > PATH_MAX)
2845 OBD_ALLOC(ldata, sizeof(*ldata));
2849 rc = linkea_data_new(ldata, &buf);
2851 GOTO(ldata_free, rc);
2853 rc = ll_getxattr(dentry, XATTR_NAME_LINK, buf.lb_buf, buf.lb_len);
2857 rc = ll_linkea_decode(ldata, linkno, &parent_fid, &ln);
2861 if (ln.ln_namelen >= name_size)
2862 GOTO(lb_free, rc = -EOVERFLOW);
2864 if (copy_to_user(&arg->gp_fid, &parent_fid, sizeof(arg->gp_fid)))
2865 GOTO(lb_free, rc = -EFAULT);
2867 if (copy_to_user(&arg->gp_name, ln.ln_name, ln.ln_namelen))
2868 GOTO(lb_free, rc = -EFAULT);
2870 if (put_user('\0', arg->gp_name + ln.ln_namelen))
2871 GOTO(lb_free, rc = -EFAULT);
2876 OBD_FREE(ldata, sizeof(*ldata));