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 atomic_set(&sbi->ll_cache.ccc_users, 0);
95 sbi->ll_cache.ccc_lru_max = lru_page_max;
96 atomic_long_set(&sbi->ll_cache.ccc_lru_left, lru_page_max);
97 spin_lock_init(&sbi->ll_cache.ccc_lru_lock);
98 INIT_LIST_HEAD(&sbi->ll_cache.ccc_lru);
100 /* turn unstable check off by default as it impacts performance */
101 sbi->ll_cache.ccc_unstable_check = 0;
102 atomic_long_set(&sbi->ll_cache.ccc_unstable_nr, 0);
103 init_waitqueue_head(&sbi->ll_cache.ccc_unstable_waitq);
105 sbi->ll_ra_info.ra_max_pages_per_file = min(pages / 32,
106 SBI_DEFAULT_READAHEAD_MAX);
107 sbi->ll_ra_info.ra_max_pages = sbi->ll_ra_info.ra_max_pages_per_file;
108 sbi->ll_ra_info.ra_max_read_ahead_whole_pages =
109 SBI_DEFAULT_READAHEAD_WHOLE_MAX;
110 INIT_LIST_HEAD(&sbi->ll_conn_chain);
111 INIT_LIST_HEAD(&sbi->ll_orphan_dentry_list);
113 ll_generate_random_uuid(uuid);
114 class_uuid_unparse(uuid, &sbi->ll_sb_uuid);
115 CDEBUG(D_CONFIG, "generated uuid: %s\n", sbi->ll_sb_uuid.uuid);
117 spin_lock(&ll_sb_lock);
118 list_add_tail(&sbi->ll_list, &ll_super_blocks);
119 spin_unlock(&ll_sb_lock);
121 sbi->ll_flags |= LL_SBI_VERBOSE;
122 #ifdef ENABLE_CHECKSUM
123 sbi->ll_flags |= LL_SBI_CHECKSUM;
126 #ifdef HAVE_LRU_RESIZE_SUPPORT
127 sbi->ll_flags |= LL_SBI_LRU_RESIZE;
130 for (i = 0; i <= LL_PROCESS_HIST_MAX; i++) {
131 spin_lock_init(&sbi->ll_rw_extents_info.pp_extents[i].
133 spin_lock_init(&sbi->ll_rw_extents_info.pp_extents[i].
137 /* metadata statahead is enabled by default */
138 sbi->ll_sa_max = LL_SA_RPC_DEF;
139 atomic_set(&sbi->ll_sa_total, 0);
140 atomic_set(&sbi->ll_sa_wrong, 0);
141 atomic_set(&sbi->ll_sa_running, 0);
142 atomic_set(&sbi->ll_agl_total, 0);
143 sbi->ll_flags |= LL_SBI_AGL_ENABLED;
146 sbi->ll_squash.rsi_uid = 0;
147 sbi->ll_squash.rsi_gid = 0;
148 INIT_LIST_HEAD(&sbi->ll_squash.rsi_nosquash_nids);
149 init_rwsem(&sbi->ll_squash.rsi_sem);
154 static void ll_free_sbi(struct super_block *sb)
156 struct ll_sb_info *sbi = ll_s2sbi(sb);
160 spin_lock(&ll_sb_lock);
161 list_del(&sbi->ll_list);
162 spin_unlock(&ll_sb_lock);
163 if (!list_empty(&sbi->ll_squash.rsi_nosquash_nids))
164 cfs_free_nidlist(&sbi->ll_squash.rsi_nosquash_nids);
165 OBD_FREE(sbi, sizeof(*sbi));
170 static int client_common_fill_super(struct super_block *sb, char *md, char *dt,
171 struct vfsmount *mnt)
173 struct inode *root = NULL;
174 struct ll_sb_info *sbi = ll_s2sbi(sb);
175 struct obd_device *obd;
176 struct obd_capa *oc = NULL;
177 struct obd_statfs *osfs = NULL;
178 struct ptlrpc_request *request = NULL;
179 struct obd_connect_data *data = NULL;
180 struct obd_uuid *uuid;
181 struct md_op_data *op_data;
182 struct lustre_md lmd;
184 int size, err, checksum;
187 obd = class_name2obd(md);
189 CERROR("MD %s: not setup or attached\n", md);
203 /* indicate the features supported by this client */
204 data->ocd_connect_flags = OBD_CONNECT_IBITS | OBD_CONNECT_NODEVOH |
205 OBD_CONNECT_ATTRFID |
206 OBD_CONNECT_VERSION | OBD_CONNECT_BRW_SIZE |
207 OBD_CONNECT_MDS_CAPA | OBD_CONNECT_OSS_CAPA |
208 OBD_CONNECT_CANCELSET | OBD_CONNECT_FID |
209 OBD_CONNECT_AT | OBD_CONNECT_LOV_V3 |
210 OBD_CONNECT_RMT_CLIENT | OBD_CONNECT_VBR |
211 OBD_CONNECT_FULL20 | OBD_CONNECT_64BITHASH|
212 OBD_CONNECT_EINPROGRESS |
213 OBD_CONNECT_JOBSTATS | OBD_CONNECT_LVB_TYPE |
214 OBD_CONNECT_LAYOUTLOCK | OBD_CONNECT_PINGLESS |
215 OBD_CONNECT_MAX_EASIZE |
216 OBD_CONNECT_FLOCK_DEAD |
217 OBD_CONNECT_DISP_STRIPE | OBD_CONNECT_LFSCK |
218 OBD_CONNECT_OPEN_BY_FID |
219 OBD_CONNECT_DIR_STRIPE;
221 if (sbi->ll_flags & LL_SBI_SOM_PREVIEW)
222 data->ocd_connect_flags |= OBD_CONNECT_SOM;
224 #ifdef HAVE_LRU_RESIZE_SUPPORT
225 if (sbi->ll_flags & LL_SBI_LRU_RESIZE)
226 data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
228 #ifdef CONFIG_FS_POSIX_ACL
229 data->ocd_connect_flags |= OBD_CONNECT_ACL | OBD_CONNECT_UMASK;
232 if (OBD_FAIL_CHECK(OBD_FAIL_MDC_LIGHTWEIGHT))
233 /* flag mdc connection as lightweight, only used for test
234 * purpose, use with care */
235 data->ocd_connect_flags |= OBD_CONNECT_LIGHTWEIGHT;
237 data->ocd_ibits_known = MDS_INODELOCK_FULL;
238 data->ocd_version = LUSTRE_VERSION_CODE;
240 if (sb->s_flags & MS_RDONLY)
241 data->ocd_connect_flags |= OBD_CONNECT_RDONLY;
242 if (sbi->ll_flags & LL_SBI_USER_XATTR)
243 data->ocd_connect_flags |= OBD_CONNECT_XATTR;
245 #ifdef HAVE_MS_FLOCK_LOCK
246 /* force vfs to use lustre handler for flock() calls - bug 10743 */
247 sb->s_flags |= MS_FLOCK_LOCK;
249 #ifdef MS_HAS_NEW_AOPS
250 sb->s_flags |= MS_HAS_NEW_AOPS;
253 if (sbi->ll_flags & LL_SBI_FLOCK)
254 sbi->ll_fop = &ll_file_operations_flock;
255 else if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
256 sbi->ll_fop = &ll_file_operations;
258 sbi->ll_fop = &ll_file_operations_noflock;
261 data->ocd_connect_flags |= OBD_CONNECT_REAL;
262 if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
263 data->ocd_connect_flags |= OBD_CONNECT_RMT_CLIENT_FORCE;
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_CACHE_SIZE);
311 obd_connect_flags2str(buf, PAGE_CACHE_SIZE,
312 valid ^ CLIENT_CONNECT_MDT_REQD, ",");
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_CACHE_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, NULL);
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_max_rw_chunk = LL_DEFAULT_MAX_RW_CHUNK;
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_RMT_CLIENT) {
360 if (!(sbi->ll_flags & LL_SBI_RMT_CLIENT)) {
361 sbi->ll_flags |= LL_SBI_RMT_CLIENT;
362 LCONSOLE_INFO("client is set as remote by default.\n");
365 if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
366 sbi->ll_flags &= ~LL_SBI_RMT_CLIENT;
367 LCONSOLE_INFO("client claims to be remote, but server "
368 "rejected, forced to be local.\n");
372 if (data->ocd_connect_flags & OBD_CONNECT_MDS_CAPA) {
373 LCONSOLE_INFO("client enabled MDS capability!\n");
374 sbi->ll_flags |= LL_SBI_MDS_CAPA;
377 if (data->ocd_connect_flags & OBD_CONNECT_OSS_CAPA) {
378 LCONSOLE_INFO("client enabled OSS capability!\n");
379 sbi->ll_flags |= LL_SBI_OSS_CAPA;
382 if (data->ocd_connect_flags & OBD_CONNECT_64BITHASH)
383 sbi->ll_flags |= LL_SBI_64BIT_HASH;
385 if (data->ocd_connect_flags & OBD_CONNECT_BRW_SIZE)
386 sbi->ll_md_brw_pages = data->ocd_brw_size >> PAGE_CACHE_SHIFT;
388 sbi->ll_md_brw_pages = 1;
390 if (data->ocd_connect_flags & OBD_CONNECT_LAYOUTLOCK)
391 sbi->ll_flags |= LL_SBI_LAYOUT_LOCK;
393 if (data->ocd_ibits_known & MDS_INODELOCK_XATTR) {
394 if (!(data->ocd_connect_flags & OBD_CONNECT_MAX_EASIZE)) {
395 LCONSOLE_INFO("%s: disabling xattr cache due to "
396 "unknown maximum xattr size.\n", dt);
398 sbi->ll_flags |= LL_SBI_XATTR_CACHE;
399 sbi->ll_xattr_cache_enabled = 1;
403 obd = class_name2obd(dt);
405 CERROR("DT %s: not setup or attached\n", dt);
406 GOTO(out_md_fid, err = -ENODEV);
409 data->ocd_connect_flags = OBD_CONNECT_GRANT | OBD_CONNECT_VERSION |
410 OBD_CONNECT_REQPORTAL | OBD_CONNECT_BRW_SIZE |
411 OBD_CONNECT_CANCELSET | OBD_CONNECT_FID |
412 OBD_CONNECT_SRVLOCK | OBD_CONNECT_TRUNCLOCK|
413 OBD_CONNECT_AT | OBD_CONNECT_RMT_CLIENT |
414 OBD_CONNECT_OSS_CAPA | OBD_CONNECT_VBR|
415 OBD_CONNECT_FULL20 | OBD_CONNECT_64BITHASH |
416 OBD_CONNECT_MAXBYTES |
417 OBD_CONNECT_EINPROGRESS |
418 OBD_CONNECT_JOBSTATS | OBD_CONNECT_LVB_TYPE |
419 OBD_CONNECT_LAYOUTLOCK |
420 OBD_CONNECT_PINGLESS | OBD_CONNECT_LFSCK;
422 if (sbi->ll_flags & LL_SBI_SOM_PREVIEW)
423 data->ocd_connect_flags |= OBD_CONNECT_SOM;
425 if (!OBD_FAIL_CHECK(OBD_FAIL_OSC_CONNECT_CKSUM)) {
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();
438 #ifdef HAVE_LRU_RESIZE_SUPPORT
439 data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
441 if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
442 data->ocd_connect_flags |= OBD_CONNECT_RMT_CLIENT_FORCE;
444 CDEBUG(D_RPCTRACE, "ocd_connect_flags: "LPX64" 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 err = obd_fid_init(sbi->ll_dt_exp->exp_obd, sbi->ll_dt_exp,
470 LUSTRE_SEQ_METADATA);
472 CERROR("%s: Can't init data layer FID infrastructure, "
473 "rc = %d\n", sbi->ll_dt_exp->exp_obd->obd_name, err);
477 mutex_lock(&sbi->ll_lco.lco_lock);
478 sbi->ll_lco.lco_flags = data->ocd_connect_flags;
479 sbi->ll_lco.lco_md_exp = sbi->ll_md_exp;
480 sbi->ll_lco.lco_dt_exp = sbi->ll_dt_exp;
481 mutex_unlock(&sbi->ll_lco.lco_lock);
483 fid_zero(&sbi->ll_root_fid);
484 err = md_getstatus(sbi->ll_md_exp, &sbi->ll_root_fid, &oc);
486 CERROR("cannot mds_connect: rc = %d\n", err);
487 GOTO(out_lock_cn_cb, err);
489 if (!fid_is_sane(&sbi->ll_root_fid)) {
490 CERROR("%s: Invalid root fid "DFID" during mount\n",
491 sbi->ll_md_exp->exp_obd->obd_name,
492 PFID(&sbi->ll_root_fid));
493 GOTO(out_lock_cn_cb, err = -EINVAL);
495 CDEBUG(D_SUPER, "rootfid "DFID"\n", PFID(&sbi->ll_root_fid));
497 sb->s_op = &lustre_super_operations;
498 #if THREAD_SIZE >= 8192 /*b=17630*/
499 sb->s_export_op = &lustre_export_operations;
503 * XXX: move this to after cbd setup? */
504 valid = OBD_MD_FLGETATTR | OBD_MD_FLBLOCKS | OBD_MD_FLMDSCAPA |
506 if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
507 valid |= OBD_MD_FLRMTPERM;
508 else if (sbi->ll_flags & LL_SBI_ACL)
509 valid |= OBD_MD_FLACL;
511 OBD_ALLOC_PTR(op_data);
513 GOTO(out_lock_cn_cb, err = -ENOMEM);
515 op_data->op_fid1 = sbi->ll_root_fid;
516 op_data->op_mode = 0;
517 op_data->op_capa1 = oc;
518 op_data->op_valid = valid;
520 err = md_getattr(sbi->ll_md_exp, op_data, &request);
523 OBD_FREE_PTR(op_data);
525 CERROR("%s: md_getattr failed for root: rc = %d\n",
526 sbi->ll_md_exp->exp_obd->obd_name, err);
527 GOTO(out_lock_cn_cb, err);
530 err = md_get_lustre_md(sbi->ll_md_exp, request, sbi->ll_dt_exp,
531 sbi->ll_md_exp, &lmd);
533 CERROR("failed to understand root inode md: rc = %d\n", err);
534 ptlrpc_req_finished(request);
535 GOTO(out_lock_cn_cb, err);
538 LASSERT(fid_is_sane(&sbi->ll_root_fid));
539 root = ll_iget(sb, cl_fid_build_ino(&sbi->ll_root_fid,
540 sbi->ll_flags & LL_SBI_32BIT_API),
542 md_free_lustre_md(sbi->ll_md_exp, &lmd);
543 ptlrpc_req_finished(request);
547 obd_free_memmd(sbi->ll_dt_exp, &lmd.lsm);
548 #ifdef CONFIG_FS_POSIX_ACL
550 posix_acl_release(lmd.posix_acl);
551 lmd.posix_acl = NULL;
554 err = IS_ERR(root) ? PTR_ERR(root) : -EBADF;
556 CERROR("lustre_lite: bad iget4 for root\n");
560 err = ll_close_thread_start(&sbi->ll_lcq);
562 CERROR("cannot start close thread: rc %d\n", err);
566 #ifdef CONFIG_FS_POSIX_ACL
567 if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
568 rct_init(&sbi->ll_rct);
569 et_init(&sbi->ll_et);
573 checksum = sbi->ll_flags & LL_SBI_CHECKSUM;
574 err = obd_set_info_async(NULL, sbi->ll_dt_exp, sizeof(KEY_CHECKSUM),
575 KEY_CHECKSUM, sizeof(checksum), &checksum,
579 err = obd_set_info_async(NULL, sbi->ll_dt_exp, sizeof(KEY_CACHE_SET),
580 KEY_CACHE_SET, sizeof(sbi->ll_cache),
581 &sbi->ll_cache, NULL);
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));
608 if (proc_lustre_fs_root != NULL) {
609 err = lprocfs_register_mountpoint(proc_lustre_fs_root, sb,
612 CERROR("%s: could not register mount in lprocfs: "
613 "rc = %d\n", ll_get_fsname(sb, NULL, 0), err);
623 obd_fid_fini(sbi->ll_dt_exp->exp_obd);
625 obd_disconnect(sbi->ll_dt_exp);
626 sbi->ll_dt_exp = NULL;
627 /* Make sure all OScs are gone, since cl_cache is accessing sbi. */
628 obd_zombie_barrier();
630 obd_fid_fini(sbi->ll_md_exp->exp_obd);
632 obd_disconnect(sbi->ll_md_exp);
633 sbi->ll_md_exp = NULL;
642 int ll_get_max_mdsize(struct ll_sb_info *sbi, int *lmmsize)
646 *lmmsize = obd_size_diskmd(sbi->ll_dt_exp, NULL);
648 rc = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_MAX_EASIZE),
649 KEY_MAX_EASIZE, &size, lmmsize, NULL);
651 CERROR("Get max mdsize error rc %d\n", rc);
656 int ll_get_default_mdsize(struct ll_sb_info *sbi, int *lmmsize)
661 rc = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_DEFAULT_EASIZE),
662 KEY_DEFAULT_EASIZE, &size, lmmsize, NULL);
664 CERROR("Get default mdsize error rc %d\n", rc);
669 int ll_get_max_cookiesize(struct ll_sb_info *sbi, int *lmmsize)
674 rc = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_MAX_COOKIESIZE),
675 KEY_MAX_COOKIESIZE, &size, lmmsize, NULL);
677 CERROR("Get max cookiesize error rc %d\n", rc);
682 int ll_get_default_cookiesize(struct ll_sb_info *sbi, int *lmmsize)
687 rc = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_DEFAULT_COOKIESIZE),
688 KEY_DEFAULT_COOKIESIZE, &size, lmmsize, NULL);
690 CERROR("Get default cookiesize error rc %d\n", rc);
695 static void ll_dump_inode(struct inode *inode)
697 struct ll_d_hlist_node *tmp;
698 int dentry_count = 0;
700 LASSERT(inode != NULL);
702 ll_d_hlist_for_each(tmp, &inode->i_dentry)
705 CERROR("%s: inode %p dump: dev=%s fid="DFID
706 " mode=%o count=%u, %d dentries\n",
707 ll_get_fsname(inode->i_sb, NULL, 0), inode,
708 ll_i2mdexp(inode)->exp_obd->obd_name, PFID(ll_inode2fid(inode)),
709 inode->i_mode, atomic_read(&inode->i_count), dentry_count);
712 void lustre_dump_dentry(struct dentry *dentry, int recur)
714 struct list_head *tmp;
717 LASSERT(dentry != NULL);
719 list_for_each(tmp, &dentry->d_subdirs)
722 CERROR("dentry %p dump: name=%.*s parent=%.*s (%p), inode=%p, count=%u,"
723 " flags=0x%x, fsdata=%p, %d subdirs\n", dentry,
724 dentry->d_name.len, dentry->d_name.name,
725 dentry->d_parent->d_name.len, dentry->d_parent->d_name.name,
726 dentry->d_parent, dentry->d_inode, ll_d_count(dentry),
727 dentry->d_flags, dentry->d_fsdata, subdirs);
728 if (dentry->d_inode != NULL)
729 ll_dump_inode(dentry->d_inode);
734 list_for_each(tmp, &dentry->d_subdirs) {
735 struct dentry *d = list_entry(tmp, struct dentry, d_u.d_child);
736 lustre_dump_dentry(d, recur - 1);
740 static void client_common_put_super(struct super_block *sb)
742 struct ll_sb_info *sbi = ll_s2sbi(sb);
745 #ifdef CONFIG_FS_POSIX_ACL
746 if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
747 et_fini(&sbi->ll_et);
748 rct_fini(&sbi->ll_rct);
752 ll_close_thread_shutdown(sbi->ll_lcq);
756 list_del(&sbi->ll_conn_chain);
758 obd_fid_fini(sbi->ll_dt_exp->exp_obd);
759 obd_disconnect(sbi->ll_dt_exp);
760 sbi->ll_dt_exp = NULL;
761 /* wait till all OSCs are gone, since cl_cache is accessing sbi.
763 obd_zombie_barrier();
765 lprocfs_unregister_mountpoint(sbi);
767 obd_fid_fini(sbi->ll_md_exp->exp_obd);
768 obd_disconnect(sbi->ll_md_exp);
769 sbi->ll_md_exp = NULL;
774 void ll_kill_super(struct super_block *sb)
776 struct ll_sb_info *sbi;
780 if (!(sb->s_flags & MS_ACTIVE))
784 /* we need restore s_dev from changed for clustred NFS before put_super
785 * because new kernels have cached s_dev and change sb->s_dev in
786 * put_super not affected real removing devices */
788 sb->s_dev = sbi->ll_sdev_orig;
789 sbi->ll_umounting = 1;
791 /* wait running statahead threads to quit */
792 while (atomic_read(&sbi->ll_sa_running) > 0)
793 schedule_timeout_and_set_state(TASK_UNINTERRUPTIBLE,
794 msecs_to_jiffies(MSEC_PER_SEC >> 3));
800 static inline int ll_set_opt(const char *opt, char *data, int fl)
802 if (strncmp(opt, data, strlen(opt)) != 0)
808 /* non-client-specific mount options are parsed in lmd_parse */
809 static int ll_options(char *options, int *flags)
812 char *s1 = options, *s2;
818 CDEBUG(D_CONFIG, "Parsing opts %s\n", options);
821 CDEBUG(D_SUPER, "next opt=%s\n", s1);
822 tmp = ll_set_opt("nolock", s1, LL_SBI_NOLCK);
827 tmp = ll_set_opt("flock", s1, LL_SBI_FLOCK);
832 tmp = ll_set_opt("localflock", s1, LL_SBI_LOCALFLOCK);
837 tmp = ll_set_opt("noflock", s1, LL_SBI_FLOCK|LL_SBI_LOCALFLOCK);
842 tmp = ll_set_opt("user_xattr", s1, LL_SBI_USER_XATTR);
847 tmp = ll_set_opt("nouser_xattr", s1, LL_SBI_USER_XATTR);
852 tmp = ll_set_opt("remote_client", s1, LL_SBI_RMT_CLIENT);
857 tmp = ll_set_opt("user_fid2path", s1, LL_SBI_USER_FID2PATH);
862 tmp = ll_set_opt("nouser_fid2path", s1, LL_SBI_USER_FID2PATH);
868 tmp = ll_set_opt("checksum", s1, LL_SBI_CHECKSUM);
873 tmp = ll_set_opt("nochecksum", s1, LL_SBI_CHECKSUM);
878 tmp = ll_set_opt("lruresize", s1, LL_SBI_LRU_RESIZE);
883 tmp = ll_set_opt("nolruresize", s1, LL_SBI_LRU_RESIZE);
888 tmp = ll_set_opt("lazystatfs", s1, LL_SBI_LAZYSTATFS);
893 tmp = ll_set_opt("nolazystatfs", s1, LL_SBI_LAZYSTATFS);
898 tmp = ll_set_opt("som_preview", s1, LL_SBI_SOM_PREVIEW);
903 tmp = ll_set_opt("32bitapi", s1, LL_SBI_32BIT_API);
908 tmp = ll_set_opt("verbose", s1, LL_SBI_VERBOSE);
913 tmp = ll_set_opt("noverbose", s1, LL_SBI_VERBOSE);
918 LCONSOLE_ERROR_MSG(0x152, "Unknown option '%s', won't mount.\n",
924 s2 = strchr(s1, ',');
932 void ll_lli_init(struct ll_inode_info *lli)
934 lli->lli_inode_magic = LLI_INODE_MAGIC;
936 lli->lli_ioepoch = 0;
937 lli->lli_maxbytes = MAX_LFS_FILESIZE;
938 spin_lock_init(&lli->lli_lock);
939 lli->lli_posix_acl = NULL;
940 lli->lli_remote_perms = NULL;
941 mutex_init(&lli->lli_rmtperm_mutex);
942 /* Do not set lli_fid, it has been initialized already. */
943 fid_zero(&lli->lli_pfid);
944 INIT_LIST_HEAD(&lli->lli_close_list);
945 INIT_LIST_HEAD(&lli->lli_oss_capas);
946 atomic_set(&lli->lli_open_count, 0);
947 lli->lli_mds_capa = NULL;
948 lli->lli_rmtperm_time = 0;
949 lli->lli_pending_och = NULL;
950 lli->lli_mds_read_och = NULL;
951 lli->lli_mds_write_och = NULL;
952 lli->lli_mds_exec_och = NULL;
953 lli->lli_open_fd_read_count = 0;
954 lli->lli_open_fd_write_count = 0;
955 lli->lli_open_fd_exec_count = 0;
956 mutex_init(&lli->lli_och_mutex);
957 spin_lock_init(&lli->lli_agl_lock);
958 lli->lli_has_smd = false;
959 spin_lock_init(&lli->lli_layout_lock);
960 ll_layout_version_set(lli, LL_LAYOUT_GEN_NONE);
961 lli->lli_clob = NULL;
963 init_rwsem(&lli->lli_xattrs_list_rwsem);
964 mutex_init(&lli->lli_xattrs_enq_lock);
966 LASSERT(lli->lli_vfs_inode.i_mode != 0);
967 if (S_ISDIR(lli->lli_vfs_inode.i_mode)) {
968 mutex_init(&lli->lli_readdir_mutex);
969 lli->lli_opendir_key = NULL;
971 spin_lock_init(&lli->lli_sa_lock);
972 lli->lli_opendir_pid = 0;
973 lli->lli_sa_enabled = 0;
975 mutex_init(&lli->lli_size_mutex);
976 lli->lli_symlink_name = NULL;
977 init_rwsem(&lli->lli_trunc_sem);
978 range_lock_tree_init(&lli->lli_write_tree);
979 init_rwsem(&lli->lli_glimpse_sem);
980 lli->lli_glimpse_time = 0;
981 INIT_LIST_HEAD(&lli->lli_agl_list);
982 lli->lli_agl_index = 0;
983 lli->lli_async_rc = 0;
985 mutex_init(&lli->lli_layout_mutex);
988 static inline int ll_bdi_register(struct backing_dev_info *bdi)
990 static atomic_t ll_bdi_num = ATOMIC_INIT(0);
992 bdi->name = "lustre";
993 return bdi_register(bdi, NULL, "lustre-%d",
994 atomic_inc_return(&ll_bdi_num));
997 int ll_fill_super(struct super_block *sb, struct vfsmount *mnt)
999 struct lustre_profile *lprof = NULL;
1000 struct lustre_sb_info *lsi = s2lsi(sb);
1001 struct ll_sb_info *sbi;
1002 char *dt = NULL, *md = NULL;
1003 char *profilenm = get_profile_name(sb);
1004 struct config_llog_instance *cfg;
1005 /* %p for void* in printf needs 16+2 characters: 0xffffffffffffffff */
1006 const int instlen = sizeof(cfg->cfg_instance) * 2 + 2;
1010 CDEBUG(D_VFSTRACE, "VFS Op: sb %p\n", sb);
1016 try_module_get(THIS_MODULE);
1018 /* client additional sb info */
1019 lsi->lsi_llsbi = sbi = ll_init_sbi();
1021 module_put(THIS_MODULE);
1026 err = ll_options(lsi->lsi_lmd->lmd_opts, &sbi->ll_flags);
1028 GOTO(out_free, err);
1030 err = bdi_init(&lsi->lsi_bdi);
1032 GOTO(out_free, err);
1033 lsi->lsi_flags |= LSI_BDI_INITIALIZED;
1034 lsi->lsi_bdi.capabilities = BDI_CAP_MAP_COPY;
1035 err = ll_bdi_register(&lsi->lsi_bdi);
1037 GOTO(out_free, err);
1039 sb->s_bdi = &lsi->lsi_bdi;
1040 #ifndef HAVE_DCACHE_LOCK
1041 /* kernel >= 2.6.38 store dentry operations in sb->s_d_op. */
1042 sb->s_d_op = &ll_d_ops;
1045 /* Generate a string unique to this super, in case some joker tries
1046 to mount the same fs at two mount points.
1047 Use the address of the super itself.*/
1048 cfg->cfg_instance = sb;
1049 cfg->cfg_uuid = lsi->lsi_llsbi->ll_sb_uuid;
1050 cfg->cfg_callback = class_config_llog_handler;
1051 /* set up client obds */
1052 err = lustre_process_log(sb, profilenm, cfg);
1054 GOTO(out_free, err);
1056 /* Profile set with LCFG_MOUNTOPT so we can find our mdc and osc obds */
1057 lprof = class_get_profile(profilenm);
1058 if (lprof == NULL) {
1059 LCONSOLE_ERROR_MSG(0x156, "The client profile '%s' could not be"
1060 " read from the MGS. Does that filesystem "
1061 "exist?\n", profilenm);
1062 GOTO(out_free, err = -EINVAL);
1064 CDEBUG(D_CONFIG, "Found profile %s: mdc=%s osc=%s\n", profilenm,
1065 lprof->lp_md, lprof->lp_dt);
1067 OBD_ALLOC(dt, strlen(lprof->lp_dt) + instlen + 2);
1069 GOTO(out_free, err = -ENOMEM);
1070 sprintf(dt, "%s-%p", lprof->lp_dt, cfg->cfg_instance);
1072 OBD_ALLOC(md, strlen(lprof->lp_md) + instlen + 2);
1074 GOTO(out_free, err = -ENOMEM);
1075 sprintf(md, "%s-%p", lprof->lp_md, cfg->cfg_instance);
1077 /* connections, registrations, sb setup */
1078 err = client_common_fill_super(sb, md, dt, mnt);
1082 OBD_FREE(md, strlen(lprof->lp_md) + instlen + 2);
1084 OBD_FREE(dt, strlen(lprof->lp_dt) + instlen + 2);
1087 else if (sbi->ll_flags & LL_SBI_VERBOSE)
1088 LCONSOLE_WARN("Mounted %s\n", profilenm);
1092 } /* ll_fill_super */
1094 void ll_put_super(struct super_block *sb)
1096 struct config_llog_instance cfg, params_cfg;
1097 struct obd_device *obd;
1098 struct lustre_sb_info *lsi = s2lsi(sb);
1099 struct ll_sb_info *sbi = ll_s2sbi(sb);
1100 char *profilenm = get_profile_name(sb);
1102 int next, force = 1, rc = 0;
1105 CDEBUG(D_VFSTRACE, "VFS Op: sb %p - %s\n", sb, profilenm);
1107 ll_print_capa_stat(sbi);
1109 cfg.cfg_instance = sb;
1110 lustre_end_log(sb, profilenm, &cfg);
1112 params_cfg.cfg_instance = sb;
1113 lustre_end_log(sb, PARAMS_FILENAME, ¶ms_cfg);
1115 if (sbi->ll_md_exp) {
1116 obd = class_exp2obd(sbi->ll_md_exp);
1118 force = obd->obd_force;
1121 /* Wait for unstable pages to be committed to stable storage */
1123 struct l_wait_info lwi = LWI_INTR(LWI_ON_SIGNAL_NOOP, NULL);
1124 rc = l_wait_event(sbi->ll_cache.ccc_unstable_waitq,
1125 atomic_long_read(&sbi->ll_cache.ccc_unstable_nr) == 0,
1129 ccc_count = atomic_long_read(&sbi->ll_cache.ccc_unstable_nr);
1130 if (force == 0 && rc != -EINTR)
1131 LASSERTF(ccc_count == 0, "count: %li\n", ccc_count);
1134 /* We need to set force before the lov_disconnect in
1135 lustre_common_put_super, since l_d cleans up osc's as well. */
1138 while ((obd = class_devices_in_group(&sbi->ll_sb_uuid,
1140 obd->obd_force = force;
1145 /* Only if client_common_fill_super succeeded */
1146 client_common_put_super(sb);
1150 while ((obd = class_devices_in_group(&sbi->ll_sb_uuid, &next)) !=NULL) {
1151 class_manual_cleanup(obd);
1154 if (sbi->ll_flags & LL_SBI_VERBOSE)
1155 LCONSOLE_WARN("Unmounted %s\n", profilenm ? profilenm : "");
1158 class_del_profile(profilenm);
1160 if (lsi->lsi_flags & LSI_BDI_INITIALIZED) {
1161 bdi_destroy(&lsi->lsi_bdi);
1162 lsi->lsi_flags &= ~LSI_BDI_INITIALIZED;
1166 lsi->lsi_llsbi = NULL;
1168 lustre_common_put_super(sb);
1170 cl_env_cache_purge(~0);
1172 module_put(THIS_MODULE);
1175 } /* client_put_super */
1177 struct inode *ll_inode_from_resource_lock(struct ldlm_lock *lock)
1179 struct inode *inode = NULL;
1181 /* NOTE: we depend on atomic igrab() -bzzz */
1182 lock_res_and_lock(lock);
1183 if (lock->l_resource->lr_lvb_inode) {
1184 struct ll_inode_info * lli;
1185 lli = ll_i2info(lock->l_resource->lr_lvb_inode);
1186 if (lli->lli_inode_magic == LLI_INODE_MAGIC) {
1187 inode = igrab(lock->l_resource->lr_lvb_inode);
1189 inode = lock->l_resource->lr_lvb_inode;
1190 LDLM_DEBUG_LIMIT(inode->i_state & I_FREEING ? D_INFO :
1191 D_WARNING, lock, "lr_lvb_inode %p is "
1192 "bogus: magic %08x",
1193 lock->l_resource->lr_lvb_inode,
1194 lli->lli_inode_magic);
1198 unlock_res_and_lock(lock);
1202 static void ll_dir_clear_lsm_md(struct inode *inode)
1204 struct ll_inode_info *lli = ll_i2info(inode);
1206 LASSERT(S_ISDIR(inode->i_mode));
1208 if (lli->lli_lsm_md != NULL) {
1209 lmv_free_memmd(lli->lli_lsm_md);
1210 lli->lli_lsm_md = NULL;
1214 static struct inode *ll_iget_anon_dir(struct super_block *sb,
1215 const struct lu_fid *fid,
1216 struct lustre_md *md)
1218 struct ll_sb_info *sbi = ll_s2sbi(sb);
1219 struct mdt_body *body = md->body;
1220 struct inode *inode;
1224 ino = cl_fid_build_ino(fid, sbi->ll_flags & LL_SBI_32BIT_API);
1225 inode = iget_locked(sb, ino);
1226 if (inode == NULL) {
1227 CERROR("%s: failed get simple inode "DFID": rc = -ENOENT\n",
1228 ll_get_fsname(sb, NULL, 0), PFID(fid));
1229 RETURN(ERR_PTR(-ENOENT));
1232 if (inode->i_state & I_NEW) {
1233 struct ll_inode_info *lli = ll_i2info(inode);
1234 struct lmv_stripe_md *lsm = md->lmv;
1236 inode->i_mode = (inode->i_mode & ~S_IFMT) |
1237 (body->mbo_mode & S_IFMT);
1238 LASSERTF(S_ISDIR(inode->i_mode), "Not slave inode "DFID"\n",
1241 LTIME_S(inode->i_mtime) = 0;
1242 LTIME_S(inode->i_atime) = 0;
1243 LTIME_S(inode->i_ctime) = 0;
1246 /* initializing backing dev info. */
1247 inode->i_mapping->backing_dev_info =
1248 &s2lsi(inode->i_sb)->lsi_bdi;
1249 inode->i_op = &ll_dir_inode_operations;
1250 inode->i_fop = &ll_dir_operations;
1251 lli->lli_fid = *fid;
1254 LASSERT(lsm != NULL);
1255 /* master object FID */
1256 lli->lli_pfid = body->mbo_fid1;
1257 CDEBUG(D_INODE, "lli %p slave "DFID" master "DFID"\n",
1258 lli, PFID(fid), PFID(&lli->lli_pfid));
1259 unlock_new_inode(inode);
1265 static int ll_init_lsm_md(struct inode *inode, struct lustre_md *md)
1268 struct lmv_stripe_md *lsm = md->lmv;
1271 LASSERT(lsm != NULL);
1272 /* XXX sigh, this lsm_root initialization should be in
1273 * LMV layer, but it needs ll_iget right now, so we
1274 * put this here right now. */
1275 for (i = 0; i < lsm->lsm_md_stripe_count; i++) {
1276 fid = &lsm->lsm_md_oinfo[i].lmo_fid;
1277 LASSERT(lsm->lsm_md_oinfo[i].lmo_root == NULL);
1278 /* Unfortunately ll_iget will call ll_update_inode,
1279 * where the initialization of slave inode is slightly
1280 * different, so it reset lsm_md to NULL to avoid
1281 * initializing lsm for slave inode. */
1282 /* For migrating inode, master stripe and master object will
1283 * be same, so we only need assign this inode */
1284 if (lsm->lsm_md_hash_type & LMV_HASH_FLAG_MIGRATION && i == 0)
1285 lsm->lsm_md_oinfo[i].lmo_root = inode;
1287 lsm->lsm_md_oinfo[i].lmo_root =
1288 ll_iget_anon_dir(inode->i_sb, fid, md);
1290 if (IS_ERR(lsm->lsm_md_oinfo[i].lmo_root)) {
1291 int rc = PTR_ERR(lsm->lsm_md_oinfo[i].lmo_root);
1293 lsm->lsm_md_oinfo[i].lmo_root = NULL;
1301 static inline int lli_lsm_md_eq(const struct lmv_stripe_md *lsm_md1,
1302 const struct lmv_stripe_md *lsm_md2)
1304 return lsm_md1->lsm_md_magic == lsm_md2->lsm_md_magic &&
1305 lsm_md1->lsm_md_stripe_count == lsm_md2->lsm_md_stripe_count &&
1306 lsm_md1->lsm_md_master_mdt_index ==
1307 lsm_md2->lsm_md_master_mdt_index &&
1308 lsm_md1->lsm_md_hash_type == lsm_md2->lsm_md_hash_type &&
1309 lsm_md1->lsm_md_layout_version ==
1310 lsm_md2->lsm_md_layout_version &&
1311 strcmp(lsm_md1->lsm_md_pool_name,
1312 lsm_md2->lsm_md_pool_name) == 0;
1315 static int ll_update_lsm_md(struct inode *inode, struct lustre_md *md)
1317 struct ll_inode_info *lli = ll_i2info(inode);
1318 struct lmv_stripe_md *lsm = md->lmv;
1322 LASSERT(S_ISDIR(inode->i_mode));
1323 CDEBUG(D_INODE, "update lsm %p of "DFID"\n", lli->lli_lsm_md,
1324 PFID(ll_inode2fid(inode)));
1326 /* no striped information from request. */
1328 if (lli->lli_lsm_md == NULL) {
1330 } else if (lli->lli_lsm_md->lsm_md_hash_type &
1331 LMV_HASH_FLAG_MIGRATION) {
1332 /* migration is done, the temporay MIGRATE layout has
1334 CDEBUG(D_INODE, DFID" finish migration.\n",
1335 PFID(ll_inode2fid(inode)));
1336 lmv_free_memmd(lli->lli_lsm_md);
1337 lli->lli_lsm_md = NULL;
1340 /* The lustre_md from req does not include stripeEA,
1341 * see ll_md_setattr */
1346 /* set the directory layout */
1347 if (lli->lli_lsm_md == NULL) {
1349 rc = ll_init_lsm_md(inode, md);
1353 lli->lli_lsm_md = lsm;
1354 /* set lsm_md to NULL, so the following free lustre_md
1355 * will not free this lsm */
1357 CDEBUG(D_INODE, "Set lsm %p magic %x to "DFID"\n", lsm,
1358 lsm->lsm_md_magic, PFID(ll_inode2fid(inode)));
1362 /* Compare the old and new stripe information */
1363 if (!lsm_md_eq(lli->lli_lsm_md, lsm)) {
1364 struct lmv_stripe_md *old_lsm = lli->lli_lsm_md;
1367 CERROR("%s: inode "DFID"(%p)'s lmv layout mismatch (%p)/(%p)"
1368 "magic:0x%x/0x%x stripe count: %d/%d master_mdt: %d/%d"
1369 "hash_type:0x%x/0x%x layout: 0x%x/0x%x pool:%s/%s\n",
1370 ll_get_fsname(inode->i_sb, NULL, 0), PFID(&lli->lli_fid),
1371 inode, lsm, old_lsm,
1372 lsm->lsm_md_magic, old_lsm->lsm_md_magic,
1373 lsm->lsm_md_stripe_count,
1374 old_lsm->lsm_md_stripe_count,
1375 lsm->lsm_md_master_mdt_index,
1376 old_lsm->lsm_md_master_mdt_index,
1377 lsm->lsm_md_hash_type, old_lsm->lsm_md_hash_type,
1378 lsm->lsm_md_layout_version,
1379 old_lsm->lsm_md_layout_version,
1380 lsm->lsm_md_pool_name,
1381 old_lsm->lsm_md_pool_name);
1383 for (idx = 0; idx < old_lsm->lsm_md_stripe_count; idx++) {
1384 CERROR("%s: sub FIDs in old lsm idx %d, old: "DFID"\n",
1385 ll_get_fsname(inode->i_sb, NULL, 0), idx,
1386 PFID(&old_lsm->lsm_md_oinfo[idx].lmo_fid));
1389 for (idx = 0; idx < lsm->lsm_md_stripe_count; idx++) {
1390 CERROR("%s: sub FIDs in new lsm idx %d, new: "DFID"\n",
1391 ll_get_fsname(inode->i_sb, NULL, 0), idx,
1392 PFID(&lsm->lsm_md_oinfo[idx].lmo_fid));
1401 void ll_clear_inode(struct inode *inode)
1403 struct ll_inode_info *lli = ll_i2info(inode);
1404 struct ll_sb_info *sbi = ll_i2sbi(inode);
1407 CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p)\n",
1408 PFID(ll_inode2fid(inode)), inode);
1410 if (S_ISDIR(inode->i_mode)) {
1411 /* these should have been cleared in ll_file_release */
1412 LASSERT(lli->lli_opendir_key == NULL);
1413 LASSERT(lli->lli_sai == NULL);
1414 LASSERT(lli->lli_opendir_pid == 0);
1417 spin_lock(&lli->lli_lock);
1418 ll_i2info(inode)->lli_flags &= ~LLIF_MDS_SIZE_LOCK;
1419 spin_unlock(&lli->lli_lock);
1420 md_null_inode(sbi->ll_md_exp, ll_inode2fid(inode));
1422 LASSERT(!lli->lli_open_fd_write_count);
1423 LASSERT(!lli->lli_open_fd_read_count);
1424 LASSERT(!lli->lli_open_fd_exec_count);
1426 if (lli->lli_mds_write_och)
1427 ll_md_real_close(inode, FMODE_WRITE);
1428 if (lli->lli_mds_exec_och)
1429 ll_md_real_close(inode, FMODE_EXEC);
1430 if (lli->lli_mds_read_och)
1431 ll_md_real_close(inode, FMODE_READ);
1433 if (S_ISLNK(inode->i_mode) && lli->lli_symlink_name) {
1434 OBD_FREE(lli->lli_symlink_name,
1435 strlen(lli->lli_symlink_name) + 1);
1436 lli->lli_symlink_name = NULL;
1439 ll_xattr_cache_destroy(inode);
1441 if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
1442 LASSERT(lli->lli_posix_acl == NULL);
1443 if (lli->lli_remote_perms) {
1444 free_rmtperm_hash(lli->lli_remote_perms);
1445 lli->lli_remote_perms = NULL;
1448 #ifdef CONFIG_FS_POSIX_ACL
1449 else if (lli->lli_posix_acl) {
1450 LASSERT(atomic_read(&lli->lli_posix_acl->a_refcount) == 1);
1451 LASSERT(lli->lli_remote_perms == NULL);
1452 posix_acl_release(lli->lli_posix_acl);
1453 lli->lli_posix_acl = NULL;
1456 lli->lli_inode_magic = LLI_INODE_DEAD;
1458 ll_clear_inode_capas(inode);
1459 if (S_ISDIR(inode->i_mode))
1460 ll_dir_clear_lsm_md(inode);
1461 else if (S_ISREG(inode->i_mode) && !is_bad_inode(inode))
1462 LASSERT(list_empty(&lli->lli_agl_list));
1465 * XXX This has to be done before lsm is freed below, because
1466 * cl_object still uses inode lsm.
1468 cl_inode_fini(inode);
1469 lli->lli_has_smd = false;
1474 static int ll_md_setattr(struct dentry *dentry, struct md_op_data *op_data,
1475 struct md_open_data **mod)
1477 struct lustre_md md;
1478 struct inode *inode = dentry->d_inode;
1479 struct ll_sb_info *sbi = ll_i2sbi(inode);
1480 struct ptlrpc_request *request = NULL;
1484 op_data = ll_prep_md_op_data(op_data, inode, NULL, NULL, 0, 0,
1485 LUSTRE_OPC_ANY, NULL);
1486 if (IS_ERR(op_data))
1487 RETURN(PTR_ERR(op_data));
1489 rc = md_setattr(sbi->ll_md_exp, op_data, NULL, 0, NULL, 0,
1492 ptlrpc_req_finished(request);
1493 if (rc == -ENOENT) {
1495 /* Unlinked special device node? Or just a race?
1496 * Pretend we done everything. */
1497 if (!S_ISREG(inode->i_mode) &&
1498 !S_ISDIR(inode->i_mode)) {
1499 ia_valid = op_data->op_attr.ia_valid;
1500 op_data->op_attr.ia_valid &= ~TIMES_SET_FLAGS;
1501 rc = simple_setattr(dentry, &op_data->op_attr);
1502 op_data->op_attr.ia_valid = ia_valid;
1504 } else if (rc != -EPERM && rc != -EACCES && rc != -ETXTBSY) {
1505 CERROR("md_setattr fails: rc = %d\n", rc);
1510 rc = md_get_lustre_md(sbi->ll_md_exp, request, sbi->ll_dt_exp,
1511 sbi->ll_md_exp, &md);
1513 ptlrpc_req_finished(request);
1517 ia_valid = op_data->op_attr.ia_valid;
1518 /* inode size will be in ll_setattr_ost, can't do it now since dirty
1519 * cache is not cleared yet. */
1520 op_data->op_attr.ia_valid &= ~(TIMES_SET_FLAGS | ATTR_SIZE);
1521 rc = simple_setattr(dentry, &op_data->op_attr);
1522 op_data->op_attr.ia_valid = ia_valid;
1524 /* Extract epoch data if obtained. */
1525 op_data->op_handle = md.body->mbo_handle;
1526 op_data->op_ioepoch = md.body->mbo_ioepoch;
1528 rc = ll_update_inode(inode, &md);
1529 ptlrpc_req_finished(request);
1534 /* Close IO epoch and send Size-on-MDS attribute update. */
1535 static int ll_setattr_done_writing(struct inode *inode,
1536 struct md_op_data *op_data,
1537 struct md_open_data *mod)
1539 struct ll_inode_info *lli = ll_i2info(inode);
1543 LASSERT(op_data != NULL);
1544 if (!S_ISREG(inode->i_mode))
1547 CDEBUG(D_INODE, "Epoch "LPU64" closed on "DFID" for truncate\n",
1548 op_data->op_ioepoch, PFID(&lli->lli_fid));
1550 op_data->op_flags = MF_EPOCH_CLOSE;
1551 ll_done_writing_attr(inode, op_data);
1552 ll_pack_inode2opdata(inode, op_data, NULL);
1554 rc = md_done_writing(ll_i2sbi(inode)->ll_md_exp, op_data, mod);
1555 if (rc == -EAGAIN) {
1556 /* MDS has instructed us to obtain Size-on-MDS attribute
1557 * from OSTs and send setattr to back to MDS. */
1558 rc = ll_som_update(inode, op_data);
1560 CERROR("%s: inode "DFID" mdc truncate failed: rc = %d\n",
1561 ll_i2sbi(inode)->ll_md_exp->exp_obd->obd_name,
1562 PFID(ll_inode2fid(inode)), rc);
1567 static int ll_setattr_ost(struct inode *inode, struct iattr *attr)
1569 struct obd_capa *capa;
1572 if (attr->ia_valid & ATTR_SIZE)
1573 capa = ll_osscapa_get(inode, CAPA_OPC_OSS_TRUNC);
1575 capa = ll_mdscapa_get(inode);
1577 rc = cl_setattr_ost(inode, attr, capa);
1579 if (attr->ia_valid & ATTR_SIZE)
1580 ll_truncate_free_capa(capa);
1587 /* If this inode has objects allocated to it (lsm != NULL), then the OST
1588 * object(s) determine the file size and mtime. Otherwise, the MDS will
1589 * keep these values until such a time that objects are allocated for it.
1590 * We do the MDS operations first, as it is checking permissions for us.
1591 * We don't to the MDS RPC if there is nothing that we want to store there,
1592 * otherwise there is no harm in updating mtime/atime on the MDS if we are
1593 * going to do an RPC anyways.
1595 * If we are doing a truncate, we will send the mtime and ctime updates
1596 * to the OST with the punch RPC, otherwise we do an explicit setattr RPC.
1597 * I don't believe it is possible to get e.g. ATTR_MTIME_SET and ATTR_SIZE
1600 * In case of HSMimport, we only set attr on MDS.
1602 int ll_setattr_raw(struct dentry *dentry, struct iattr *attr, bool hsm_import)
1604 struct inode *inode = dentry->d_inode;
1605 struct ll_inode_info *lli = ll_i2info(inode);
1606 struct md_op_data *op_data = NULL;
1607 struct md_open_data *mod = NULL;
1608 bool file_is_released = false;
1609 int rc = 0, rc1 = 0;
1612 CDEBUG(D_VFSTRACE, "%s: setattr inode "DFID"(%p) from %llu to %llu, "
1613 "valid %x, hsm_import %d\n",
1614 ll_get_fsname(inode->i_sb, NULL, 0), PFID(&lli->lli_fid),
1615 inode, i_size_read(inode), attr->ia_size, attr->ia_valid,
1618 if (attr->ia_valid & ATTR_SIZE) {
1619 /* Check new size against VFS/VM file size limit and rlimit */
1620 rc = inode_newsize_ok(inode, attr->ia_size);
1624 /* The maximum Lustre file size is variable, based on the
1625 * OST maximum object size and number of stripes. This
1626 * needs another check in addition to the VFS check above. */
1627 if (attr->ia_size > ll_file_maxbytes(inode)) {
1628 CDEBUG(D_INODE,"file "DFID" too large %llu > "LPU64"\n",
1629 PFID(&lli->lli_fid), attr->ia_size,
1630 ll_file_maxbytes(inode));
1634 attr->ia_valid |= ATTR_MTIME | ATTR_CTIME;
1637 /* POSIX: check before ATTR_*TIME_SET set (from inode_change_ok) */
1638 if (attr->ia_valid & TIMES_SET_FLAGS) {
1639 if ((!uid_eq(current_fsuid(), inode->i_uid)) &&
1640 !cfs_capable(CFS_CAP_FOWNER))
1644 /* We mark all of the fields "set" so MDS/OST does not re-set them */
1645 if (attr->ia_valid & ATTR_CTIME) {
1646 attr->ia_ctime = CFS_CURRENT_TIME;
1647 attr->ia_valid |= ATTR_CTIME_SET;
1649 if (!(attr->ia_valid & ATTR_ATIME_SET) &&
1650 (attr->ia_valid & ATTR_ATIME)) {
1651 attr->ia_atime = CFS_CURRENT_TIME;
1652 attr->ia_valid |= ATTR_ATIME_SET;
1654 if (!(attr->ia_valid & ATTR_MTIME_SET) &&
1655 (attr->ia_valid & ATTR_MTIME)) {
1656 attr->ia_mtime = CFS_CURRENT_TIME;
1657 attr->ia_valid |= ATTR_MTIME_SET;
1660 if (attr->ia_valid & (ATTR_MTIME | ATTR_CTIME))
1661 CDEBUG(D_INODE, "setting mtime %lu, ctime %lu, now = %lu\n",
1662 LTIME_S(attr->ia_mtime), LTIME_S(attr->ia_ctime),
1663 cfs_time_current_sec());
1665 /* We always do an MDS RPC, even if we're only changing the size;
1666 * only the MDS knows whether truncate() should fail with -ETXTBUSY */
1668 OBD_ALLOC_PTR(op_data);
1669 if (op_data == NULL)
1672 if (!S_ISDIR(inode->i_mode)) {
1673 if (attr->ia_valid & ATTR_SIZE)
1674 inode_dio_write_done(inode);
1675 mutex_unlock(&inode->i_mutex);
1678 /* truncate on a released file must failed with -ENODATA,
1679 * so size must not be set on MDS for released file
1680 * but other attributes must be set
1682 if (S_ISREG(inode->i_mode)) {
1683 struct lov_stripe_md *lsm;
1686 ll_layout_refresh(inode, &gen);
1687 lsm = ccc_inode_lsm_get(inode);
1688 if (lsm && lsm->lsm_pattern & LOV_PATTERN_F_RELEASED)
1689 file_is_released = true;
1690 ccc_inode_lsm_put(inode, lsm);
1692 if (!hsm_import && attr->ia_valid & ATTR_SIZE) {
1693 if (file_is_released) {
1694 rc = ll_layout_restore(inode, 0, attr->ia_size);
1698 file_is_released = false;
1699 ll_layout_refresh(inode, &gen);
1702 /* If we are changing file size, file content is
1703 * modified, flag it. */
1704 attr->ia_valid |= MDS_OPEN_OWNEROVERRIDE;
1705 spin_lock(&lli->lli_lock);
1706 lli->lli_flags |= LLIF_DATA_MODIFIED;
1707 spin_unlock(&lli->lli_lock);
1708 op_data->op_bias |= MDS_DATA_MODIFIED;
1712 memcpy(&op_data->op_attr, attr, sizeof(*attr));
1714 /* Open epoch for truncate. */
1715 if (exp_connect_som(ll_i2mdexp(inode)) && !hsm_import &&
1716 (attr->ia_valid & (ATTR_SIZE | ATTR_MTIME | ATTR_MTIME_SET)))
1717 op_data->op_flags = MF_EPOCH_OPEN;
1719 rc = ll_md_setattr(dentry, op_data, &mod);
1723 /* RPC to MDT is sent, cancel data modification flag */
1724 if (rc == 0 && (op_data->op_bias & MDS_DATA_MODIFIED)) {
1725 spin_lock(&lli->lli_lock);
1726 lli->lli_flags &= ~LLIF_DATA_MODIFIED;
1727 spin_unlock(&lli->lli_lock);
1730 ll_ioepoch_open(lli, op_data->op_ioepoch);
1731 if (!S_ISREG(inode->i_mode) || file_is_released)
1734 if (attr->ia_valid & (ATTR_SIZE |
1735 ATTR_ATIME | ATTR_ATIME_SET |
1736 ATTR_MTIME | ATTR_MTIME_SET)) {
1737 /* For truncate and utimes sending attributes to OSTs, setting
1738 * mtime/atime to the past will be performed under PW [0:EOF]
1739 * extent lock (new_size:EOF for truncate). It may seem
1740 * excessive to send mtime/atime updates to OSTs when not
1741 * setting times to past, but it is necessary due to possible
1742 * time de-synchronization between MDT inode and OST objects */
1743 if (attr->ia_valid & ATTR_SIZE)
1744 down_write(&lli->lli_trunc_sem);
1745 rc = ll_setattr_ost(inode, attr);
1746 if (attr->ia_valid & ATTR_SIZE)
1747 up_write(&lli->lli_trunc_sem);
1752 if (op_data->op_ioepoch) {
1753 rc1 = ll_setattr_done_writing(inode, op_data, mod);
1757 ll_finish_md_op_data(op_data);
1759 if (!S_ISDIR(inode->i_mode)) {
1760 mutex_lock(&inode->i_mutex);
1761 if ((attr->ia_valid & ATTR_SIZE) && !hsm_import)
1762 inode_dio_wait(inode);
1765 ll_stats_ops_tally(ll_i2sbi(inode), (attr->ia_valid & ATTR_SIZE) ?
1766 LPROC_LL_TRUNC : LPROC_LL_SETATTR, 1);
1771 int ll_setattr(struct dentry *de, struct iattr *attr)
1773 int mode = de->d_inode->i_mode;
1775 if ((attr->ia_valid & (ATTR_CTIME|ATTR_SIZE|ATTR_MODE)) ==
1776 (ATTR_CTIME|ATTR_SIZE|ATTR_MODE))
1777 attr->ia_valid |= MDS_OPEN_OWNEROVERRIDE;
1779 if (((attr->ia_valid & (ATTR_MODE|ATTR_FORCE|ATTR_SIZE)) ==
1780 (ATTR_SIZE|ATTR_MODE)) &&
1781 (((mode & S_ISUID) && !(attr->ia_mode & S_ISUID)) ||
1782 (((mode & (S_ISGID|S_IXGRP)) == (S_ISGID|S_IXGRP)) &&
1783 !(attr->ia_mode & S_ISGID))))
1784 attr->ia_valid |= ATTR_FORCE;
1786 if ((attr->ia_valid & ATTR_MODE) &&
1788 !(attr->ia_mode & S_ISUID) &&
1789 !(attr->ia_valid & ATTR_KILL_SUID))
1790 attr->ia_valid |= ATTR_KILL_SUID;
1792 if ((attr->ia_valid & ATTR_MODE) &&
1793 ((mode & (S_ISGID|S_IXGRP)) == (S_ISGID|S_IXGRP)) &&
1794 !(attr->ia_mode & S_ISGID) &&
1795 !(attr->ia_valid & ATTR_KILL_SGID))
1796 attr->ia_valid |= ATTR_KILL_SGID;
1798 return ll_setattr_raw(de, attr, false);
1801 int ll_statfs_internal(struct super_block *sb, struct obd_statfs *osfs,
1802 __u64 max_age, __u32 flags)
1804 struct ll_sb_info *sbi = ll_s2sbi(sb);
1805 struct obd_statfs obd_osfs;
1809 rc = obd_statfs(NULL, sbi->ll_md_exp, osfs, max_age, flags);
1811 CERROR("md_statfs fails: rc = %d\n", rc);
1815 osfs->os_type = sb->s_magic;
1817 CDEBUG(D_SUPER, "MDC blocks "LPU64"/"LPU64" objects "LPU64"/"LPU64"\n",
1818 osfs->os_bavail, osfs->os_blocks, osfs->os_ffree,osfs->os_files);
1820 if (sbi->ll_flags & LL_SBI_LAZYSTATFS)
1821 flags |= OBD_STATFS_NODELAY;
1823 rc = obd_statfs_rqset(sbi->ll_dt_exp, &obd_osfs, max_age, flags);
1825 CERROR("obd_statfs fails: rc = %d\n", rc);
1829 CDEBUG(D_SUPER, "OSC blocks "LPU64"/"LPU64" objects "LPU64"/"LPU64"\n",
1830 obd_osfs.os_bavail, obd_osfs.os_blocks, obd_osfs.os_ffree,
1833 osfs->os_bsize = obd_osfs.os_bsize;
1834 osfs->os_blocks = obd_osfs.os_blocks;
1835 osfs->os_bfree = obd_osfs.os_bfree;
1836 osfs->os_bavail = obd_osfs.os_bavail;
1838 /* If we don't have as many objects free on the OST as inodes
1839 * on the MDS, we reduce the total number of inodes to
1840 * compensate, so that the "inodes in use" number is correct.
1842 if (obd_osfs.os_ffree < osfs->os_ffree) {
1843 osfs->os_files = (osfs->os_files - osfs->os_ffree) +
1845 osfs->os_ffree = obd_osfs.os_ffree;
1850 int ll_statfs(struct dentry *de, struct kstatfs *sfs)
1852 struct super_block *sb = de->d_sb;
1853 struct obd_statfs osfs;
1854 __u64 fsid = huge_encode_dev(sb->s_dev);
1857 CDEBUG(D_VFSTRACE, "VFS Op: at "LPU64" jiffies\n", get_jiffies_64());
1858 ll_stats_ops_tally(ll_s2sbi(sb), LPROC_LL_STAFS, 1);
1860 /* Some amount of caching on the client is allowed */
1861 rc = ll_statfs_internal(sb, &osfs,
1862 cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS),
1867 statfs_unpack(sfs, &osfs);
1869 /* We need to downshift for all 32-bit kernels, because we can't
1870 * tell if the kernel is being called via sys_statfs64() or not.
1871 * Stop before overflowing f_bsize - in which case it is better
1872 * to just risk EOVERFLOW if caller is using old sys_statfs(). */
1873 if (sizeof(long) < 8) {
1874 while (osfs.os_blocks > ~0UL && sfs->f_bsize < 0x40000000) {
1877 osfs.os_blocks >>= 1;
1878 osfs.os_bfree >>= 1;
1879 osfs.os_bavail >>= 1;
1883 sfs->f_blocks = osfs.os_blocks;
1884 sfs->f_bfree = osfs.os_bfree;
1885 sfs->f_bavail = osfs.os_bavail;
1886 sfs->f_fsid.val[0] = (__u32)fsid;
1887 sfs->f_fsid.val[1] = (__u32)(fsid >> 32);
1891 void ll_inode_size_lock(struct inode *inode)
1893 struct ll_inode_info *lli;
1895 LASSERT(!S_ISDIR(inode->i_mode));
1897 lli = ll_i2info(inode);
1898 mutex_lock(&lli->lli_size_mutex);
1901 void ll_inode_size_unlock(struct inode *inode)
1903 struct ll_inode_info *lli;
1905 lli = ll_i2info(inode);
1906 mutex_unlock(&lli->lli_size_mutex);
1909 int ll_update_inode(struct inode *inode, struct lustre_md *md)
1911 struct ll_inode_info *lli = ll_i2info(inode);
1912 struct mdt_body *body = md->body;
1913 struct lov_stripe_md *lsm = md->lsm;
1914 struct ll_sb_info *sbi = ll_i2sbi(inode);
1916 LASSERT((lsm != NULL) == ((body->mbo_valid & OBD_MD_FLEASIZE) != 0));
1918 if (!lli->lli_has_smd &&
1919 !(sbi->ll_flags & LL_SBI_LAYOUT_LOCK))
1920 cl_file_inode_init(inode, md);
1922 lli->lli_maxbytes = lsm->lsm_maxbytes;
1923 if (lli->lli_maxbytes > MAX_LFS_FILESIZE)
1924 lli->lli_maxbytes = MAX_LFS_FILESIZE;
1927 if (S_ISDIR(inode->i_mode)) {
1930 rc = ll_update_lsm_md(inode, md);
1935 if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
1936 if (body->mbo_valid & OBD_MD_FLRMTPERM)
1937 ll_update_remote_perm(inode, md->remote_perm);
1939 #ifdef CONFIG_FS_POSIX_ACL
1940 else if (body->mbo_valid & OBD_MD_FLACL) {
1941 spin_lock(&lli->lli_lock);
1942 if (lli->lli_posix_acl)
1943 posix_acl_release(lli->lli_posix_acl);
1944 lli->lli_posix_acl = md->posix_acl;
1945 spin_unlock(&lli->lli_lock);
1948 inode->i_ino = cl_fid_build_ino(&body->mbo_fid1,
1949 sbi->ll_flags & LL_SBI_32BIT_API);
1950 inode->i_generation = cl_fid_build_gen(&body->mbo_fid1);
1952 if (body->mbo_valid & OBD_MD_FLATIME) {
1953 if (body->mbo_atime > LTIME_S(inode->i_atime))
1954 LTIME_S(inode->i_atime) = body->mbo_atime;
1955 lli->lli_atime = body->mbo_atime;
1958 if (body->mbo_valid & OBD_MD_FLMTIME) {
1959 if (body->mbo_mtime > LTIME_S(inode->i_mtime)) {
1960 CDEBUG(D_INODE, "setting ino %lu mtime from %lu "
1961 "to "LPU64"\n", inode->i_ino,
1962 LTIME_S(inode->i_mtime), body->mbo_mtime);
1963 LTIME_S(inode->i_mtime) = body->mbo_mtime;
1965 lli->lli_mtime = body->mbo_mtime;
1968 if (body->mbo_valid & OBD_MD_FLCTIME) {
1969 if (body->mbo_ctime > LTIME_S(inode->i_ctime))
1970 LTIME_S(inode->i_ctime) = body->mbo_ctime;
1971 lli->lli_ctime = body->mbo_ctime;
1974 if (body->mbo_valid & OBD_MD_FLMODE)
1975 inode->i_mode = (inode->i_mode & S_IFMT) |
1976 (body->mbo_mode & ~S_IFMT);
1978 if (body->mbo_valid & OBD_MD_FLTYPE)
1979 inode->i_mode = (inode->i_mode & ~S_IFMT) |
1980 (body->mbo_mode & S_IFMT);
1982 LASSERT(inode->i_mode != 0);
1983 if (S_ISREG(inode->i_mode))
1984 inode->i_blkbits = min(PTLRPC_MAX_BRW_BITS + 1,
1985 LL_MAX_BLKSIZE_BITS);
1987 inode->i_blkbits = inode->i_sb->s_blocksize_bits;
1989 if (body->mbo_valid & OBD_MD_FLUID)
1990 inode->i_uid = make_kuid(&init_user_ns, body->mbo_uid);
1991 if (body->mbo_valid & OBD_MD_FLGID)
1992 inode->i_gid = make_kgid(&init_user_ns, body->mbo_gid);
1993 if (body->mbo_valid & OBD_MD_FLFLAGS)
1994 inode->i_flags = ll_ext_to_inode_flags(body->mbo_flags);
1995 if (body->mbo_valid & OBD_MD_FLNLINK)
1996 set_nlink(inode, body->mbo_nlink);
1997 if (body->mbo_valid & OBD_MD_FLRDEV)
1998 inode->i_rdev = old_decode_dev(body->mbo_rdev);
2000 if (body->mbo_valid & OBD_MD_FLID) {
2001 /* FID shouldn't be changed! */
2002 if (fid_is_sane(&lli->lli_fid)) {
2003 LASSERTF(lu_fid_eq(&lli->lli_fid, &body->mbo_fid1),
2004 "Trying to change FID "DFID
2005 " to the "DFID", inode "DFID"(%p)\n",
2006 PFID(&lli->lli_fid), PFID(&body->mbo_fid1),
2007 PFID(ll_inode2fid(inode)), inode);
2009 lli->lli_fid = body->mbo_fid1;
2013 LASSERT(fid_seq(&lli->lli_fid) != 0);
2015 if (body->mbo_valid & OBD_MD_FLSIZE) {
2016 if (exp_connect_som(ll_i2mdexp(inode)) &&
2017 S_ISREG(inode->i_mode)) {
2018 struct lustre_handle lockh;
2021 /* As it is possible a blocking ast has been processed
2022 * by this time, we need to check there is an UPDATE
2023 * lock on the client and set LLIF_MDS_SIZE_LOCK holding
2025 mode = ll_take_md_lock(inode, MDS_INODELOCK_UPDATE,
2026 &lockh, LDLM_FL_CBPENDING,
2030 if (lli->lli_flags & (LLIF_DONE_WRITING |
2031 LLIF_EPOCH_PENDING |
2033 CERROR("%s: inode "DFID" flags %u still"
2034 " has size authority! do not "
2035 "trust the size from MDS\n",
2036 sbi->ll_md_exp->exp_obd->obd_name,
2037 PFID(ll_inode2fid(inode)),
2040 /* Use old size assignment to avoid
2041 * deadlock bz14138 & bz14326 */
2042 i_size_write(inode, body->mbo_size);
2043 spin_lock(&lli->lli_lock);
2044 lli->lli_flags |= LLIF_MDS_SIZE_LOCK;
2045 spin_unlock(&lli->lli_lock);
2047 ldlm_lock_decref(&lockh, mode);
2050 /* Use old size assignment to avoid
2051 * deadlock bz14138 & bz14326 */
2052 i_size_write(inode, body->mbo_size);
2055 "inode="DFID", updating i_size %llu\n",
2056 PFID(ll_inode2fid(inode)),
2057 (unsigned long long)body->mbo_size);
2060 if (body->mbo_valid & OBD_MD_FLBLOCKS)
2061 inode->i_blocks = body->mbo_blocks;
2064 if (body->mbo_valid & OBD_MD_FLMDSCAPA) {
2065 LASSERT(md->mds_capa);
2066 ll_add_capa(inode, md->mds_capa);
2069 if (body->mbo_valid & OBD_MD_FLOSSCAPA) {
2070 LASSERT(md->oss_capa);
2071 ll_add_capa(inode, md->oss_capa);
2074 if (body->mbo_valid & OBD_MD_TSTATE) {
2075 if (body->mbo_t_state & MS_RESTORE)
2076 lli->lli_flags |= LLIF_FILE_RESTORING;
2082 int ll_read_inode2(struct inode *inode, void *opaque)
2084 struct lustre_md *md = opaque;
2085 struct ll_inode_info *lli = ll_i2info(inode);
2089 CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p)\n",
2090 PFID(&lli->lli_fid), inode);
2092 LASSERT(!lli->lli_has_smd);
2094 /* Core attributes from the MDS first. This is a new inode, and
2095 * the VFS doesn't zero times in the core inode so we have to do
2096 * it ourselves. They will be overwritten by either MDS or OST
2097 * attributes - we just need to make sure they aren't newer. */
2098 LTIME_S(inode->i_mtime) = 0;
2099 LTIME_S(inode->i_atime) = 0;
2100 LTIME_S(inode->i_ctime) = 0;
2102 rc = ll_update_inode(inode, md);
2106 /* OIDEBUG(inode); */
2108 /* initializing backing dev info. */
2109 inode->i_mapping->backing_dev_info = &s2lsi(inode->i_sb)->lsi_bdi;
2112 if (S_ISREG(inode->i_mode)) {
2113 struct ll_sb_info *sbi = ll_i2sbi(inode);
2114 inode->i_op = &ll_file_inode_operations;
2115 inode->i_fop = sbi->ll_fop;
2116 inode->i_mapping->a_ops = (struct address_space_operations *)&ll_aops;
2118 } else if (S_ISDIR(inode->i_mode)) {
2119 inode->i_op = &ll_dir_inode_operations;
2120 inode->i_fop = &ll_dir_operations;
2122 } else if (S_ISLNK(inode->i_mode)) {
2123 inode->i_op = &ll_fast_symlink_inode_operations;
2126 inode->i_op = &ll_special_inode_operations;
2128 init_special_inode(inode, inode->i_mode,
2137 void ll_delete_inode(struct inode *inode)
2139 struct ll_inode_info *lli = ll_i2info(inode);
2142 if (S_ISREG(inode->i_mode) && lli->lli_clob != NULL)
2143 /* It is last chance to write out dirty pages,
2144 * otherwise we may lose data while umount */
2145 cl_sync_file_range(inode, 0, OBD_OBJECT_EOF, CL_FSYNC_LOCAL, 1);
2147 truncate_inode_pages_final(&inode->i_data);
2149 LASSERTF(inode->i_data.nrpages == 0, "inode="DFID"(%p) nrpages=%lu, "
2150 "see https://jira.hpdd.intel.com/browse/LU-118\n",
2151 PFID(ll_inode2fid(inode)), inode, inode->i_data.nrpages);
2153 #ifdef HAVE_SBOPS_EVICT_INODE
2154 ll_clear_inode(inode);
2161 int ll_iocontrol(struct inode *inode, struct file *file,
2162 unsigned int cmd, unsigned long arg)
2164 struct ll_sb_info *sbi = ll_i2sbi(inode);
2165 struct ptlrpc_request *req = NULL;
2170 case FSFILT_IOC_GETFLAGS: {
2171 struct mdt_body *body;
2172 struct md_op_data *op_data;
2174 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL,
2175 0, 0, LUSTRE_OPC_ANY,
2177 if (IS_ERR(op_data))
2178 RETURN(PTR_ERR(op_data));
2180 op_data->op_valid = OBD_MD_FLFLAGS;
2181 rc = md_getattr(sbi->ll_md_exp, op_data, &req);
2182 ll_finish_md_op_data(op_data);
2184 CERROR("%s: failure inode "DFID": rc = %d\n",
2185 sbi->ll_md_exp->exp_obd->obd_name,
2186 PFID(ll_inode2fid(inode)), rc);
2190 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
2192 flags = body->mbo_flags;
2194 ptlrpc_req_finished(req);
2196 RETURN(put_user(flags, (int __user *)arg));
2198 case FSFILT_IOC_SETFLAGS: {
2199 struct lov_stripe_md *lsm;
2200 struct obd_info oinfo = { { { 0 } } };
2201 struct md_op_data *op_data;
2203 if (get_user(flags, (int __user *)arg))
2206 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
2207 LUSTRE_OPC_ANY, NULL);
2208 if (IS_ERR(op_data))
2209 RETURN(PTR_ERR(op_data));
2211 op_data->op_attr_flags = flags;
2212 op_data->op_attr.ia_valid |= ATTR_ATTR_FLAG;
2213 rc = md_setattr(sbi->ll_md_exp, op_data,
2214 NULL, 0, NULL, 0, &req, NULL);
2215 ll_finish_md_op_data(op_data);
2216 ptlrpc_req_finished(req);
2220 inode->i_flags = ll_ext_to_inode_flags(flags);
2222 lsm = ccc_inode_lsm_get(inode);
2223 if (!lsm_has_objects(lsm)) {
2224 ccc_inode_lsm_put(inode, lsm);
2228 OBDO_ALLOC(oinfo.oi_oa);
2230 ccc_inode_lsm_put(inode, lsm);
2234 oinfo.oi_oa->o_oi = lsm->lsm_oi;
2235 oinfo.oi_oa->o_flags = flags;
2236 oinfo.oi_oa->o_valid = OBD_MD_FLID | OBD_MD_FLFLAGS |
2238 oinfo.oi_capa = ll_mdscapa_get(inode);
2239 obdo_set_parent_fid(oinfo.oi_oa, &ll_i2info(inode)->lli_fid);
2240 rc = obd_setattr_rqset(sbi->ll_dt_exp, &oinfo, NULL);
2241 capa_put(oinfo.oi_capa);
2242 OBDO_FREE(oinfo.oi_oa);
2243 ccc_inode_lsm_put(inode, lsm);
2245 if (rc && rc != -EPERM && rc != -EACCES)
2246 CERROR("osc_setattr_async fails: rc = %d\n", rc);
2257 int ll_flush_ctx(struct inode *inode)
2259 struct ll_sb_info *sbi = ll_i2sbi(inode);
2261 CDEBUG(D_SEC, "flush context for user %d\n",
2262 from_kuid(&init_user_ns, current_uid()));
2264 obd_set_info_async(NULL, sbi->ll_md_exp,
2265 sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
2267 obd_set_info_async(NULL, sbi->ll_dt_exp,
2268 sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
2273 /* umount -f client means force down, don't save state */
2274 void ll_umount_begin(struct super_block *sb)
2276 struct ll_sb_info *sbi = ll_s2sbi(sb);
2277 struct obd_device *obd;
2278 struct obd_ioctl_data *ioc_data;
2281 CDEBUG(D_VFSTRACE, "VFS Op: superblock %p count %d active %d\n", sb,
2282 sb->s_count, atomic_read(&sb->s_active));
2284 obd = class_exp2obd(sbi->ll_md_exp);
2286 CERROR("Invalid MDC connection handle "LPX64"\n",
2287 sbi->ll_md_exp->exp_handle.h_cookie);
2293 obd = class_exp2obd(sbi->ll_dt_exp);
2295 CERROR("Invalid LOV connection handle "LPX64"\n",
2296 sbi->ll_dt_exp->exp_handle.h_cookie);
2302 OBD_ALLOC_PTR(ioc_data);
2304 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_md_exp,
2305 sizeof *ioc_data, ioc_data, NULL);
2307 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_dt_exp,
2308 sizeof *ioc_data, ioc_data, NULL);
2310 OBD_FREE_PTR(ioc_data);
2313 /* Really, we'd like to wait until there are no requests outstanding,
2314 * and then continue. For now, we just invalidate the requests,
2315 * schedule() and sleep one second if needed, and hope.
2321 int ll_remount_fs(struct super_block *sb, int *flags, char *data)
2323 struct ll_sb_info *sbi = ll_s2sbi(sb);
2324 char *profilenm = get_profile_name(sb);
2328 if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY)) {
2329 read_only = *flags & MS_RDONLY;
2330 err = obd_set_info_async(NULL, sbi->ll_md_exp,
2331 sizeof(KEY_READ_ONLY),
2332 KEY_READ_ONLY, sizeof(read_only),
2335 LCONSOLE_WARN("Failed to remount %s %s (%d)\n",
2336 profilenm, read_only ?
2337 "read-only" : "read-write", err);
2342 sb->s_flags |= MS_RDONLY;
2344 sb->s_flags &= ~MS_RDONLY;
2346 if (sbi->ll_flags & LL_SBI_VERBOSE)
2347 LCONSOLE_WARN("Remounted %s %s\n", profilenm,
2348 read_only ? "read-only" : "read-write");
2353 int ll_prep_inode(struct inode **inode, struct ptlrpc_request *req,
2354 struct super_block *sb, struct lookup_intent *it)
2356 struct ll_sb_info *sbi = NULL;
2357 struct lustre_md md = { NULL };
2361 LASSERT(*inode || sb);
2362 sbi = sb ? ll_s2sbi(sb) : ll_i2sbi(*inode);
2363 rc = md_get_lustre_md(sbi->ll_md_exp, req, sbi->ll_dt_exp,
2364 sbi->ll_md_exp, &md);
2369 rc = ll_update_inode(*inode, &md);
2373 LASSERT(sb != NULL);
2376 * At this point server returns to client's same fid as client
2377 * generated for creating. So using ->fid1 is okay here.
2379 LASSERT(fid_is_sane(&md.body->mbo_fid1));
2381 *inode = ll_iget(sb, cl_fid_build_ino(&md.body->mbo_fid1,
2382 sbi->ll_flags & LL_SBI_32BIT_API),
2384 if (IS_ERR(*inode)) {
2385 #ifdef CONFIG_FS_POSIX_ACL
2387 posix_acl_release(md.posix_acl);
2388 md.posix_acl = NULL;
2391 rc = IS_ERR(*inode) ? PTR_ERR(*inode) : -ENOMEM;
2393 CERROR("new_inode -fatal: rc %d\n", rc);
2398 /* Handling piggyback layout lock.
2399 * Layout lock can be piggybacked by getattr and open request.
2400 * The lsm can be applied to inode only if it comes with a layout lock
2401 * otherwise correct layout may be overwritten, for example:
2402 * 1. proc1: mdt returns a lsm but not granting layout
2403 * 2. layout was changed by another client
2404 * 3. proc2: refresh layout and layout lock granted
2405 * 4. proc1: to apply a stale layout */
2406 if (it != NULL && it->d.lustre.it_lock_mode != 0) {
2407 struct lustre_handle lockh;
2408 struct ldlm_lock *lock;
2410 lockh.cookie = it->d.lustre.it_lock_handle;
2411 lock = ldlm_handle2lock(&lockh);
2412 LASSERT(lock != NULL);
2413 if (ldlm_has_layout(lock)) {
2414 struct cl_object_conf conf;
2416 memset(&conf, 0, sizeof(conf));
2417 conf.coc_opc = OBJECT_CONF_SET;
2418 conf.coc_inode = *inode;
2419 conf.coc_lock = lock;
2420 conf.u.coc_md = &md;
2421 (void)ll_layout_conf(*inode, &conf);
2423 LDLM_LOCK_PUT(lock);
2428 obd_free_memmd(sbi->ll_dt_exp, &md.lsm);
2429 md_free_lustre_md(sbi->ll_md_exp, &md);
2433 int ll_obd_statfs(struct inode *inode, void __user *arg)
2435 struct ll_sb_info *sbi = NULL;
2436 struct obd_export *exp;
2438 struct obd_ioctl_data *data = NULL;
2440 __u32 __user flags; /* not user, but obd_iocontrol is abused */
2443 if (!inode || !(sbi = ll_i2sbi(inode)))
2444 GOTO(out_statfs, rc = -EINVAL);
2446 rc = obd_ioctl_getdata(&buf, &len, arg);
2448 GOTO(out_statfs, rc);
2451 if (!data->ioc_inlbuf1 || !data->ioc_inlbuf2 ||
2452 !data->ioc_pbuf1 || !data->ioc_pbuf2)
2453 GOTO(out_statfs, rc = -EINVAL);
2455 if (data->ioc_inllen1 != sizeof(__u32) ||
2456 data->ioc_inllen2 != sizeof(__u32) ||
2457 data->ioc_plen1 != sizeof(struct obd_statfs) ||
2458 data->ioc_plen2 != sizeof(struct obd_uuid))
2459 GOTO(out_statfs, rc = -EINVAL);
2461 memcpy(&type, data->ioc_inlbuf1, sizeof(__u32));
2462 if (type & LL_STATFS_LMV)
2463 exp = sbi->ll_md_exp;
2464 else if (type & LL_STATFS_LOV)
2465 exp = sbi->ll_dt_exp;
2467 GOTO(out_statfs, rc = -ENODEV);
2469 flags = (type & LL_STATFS_NODELAY) ? OBD_STATFS_NODELAY : 0;
2470 rc = obd_iocontrol(IOC_OBD_STATFS, exp, len, buf, &flags);
2472 GOTO(out_statfs, rc);
2475 obd_ioctl_freedata(buf, len);
2479 int ll_process_config(struct lustre_cfg *lcfg)
2481 struct super_block *sb;
2486 /* The instance name contains the sb: lustre-client-aacfe000 */
2487 ptr = strrchr(lustre_cfg_string(lcfg, 0), '-');
2488 if (!ptr || !*(++ptr))
2490 if (sscanf(ptr, "%lx", &x) != 1)
2492 sb = (struct super_block *)x;
2493 /* This better be a real Lustre superblock! */
2494 LASSERT(s2lsi(sb)->lsi_lmd->lmd_magic == LMD_MAGIC);
2496 /* Note we have not called client_common_fill_super yet, so
2497 proc fns must be able to handle that! */
2498 rc = class_process_proc_param(PARAM_LLITE, lprocfs_llite_obd_vars,
2505 /* this function prepares md_op_data hint for passing ot down to MD stack. */
2506 struct md_op_data * ll_prep_md_op_data(struct md_op_data *op_data,
2507 struct inode *i1, struct inode *i2,
2508 const char *name, size_t namelen,
2509 __u32 mode, __u32 opc, void *data)
2511 LASSERT(i1 != NULL);
2514 /* Do not reuse namelen for something else. */
2516 return ERR_PTR(-EINVAL);
2518 if (namelen > ll_i2sbi(i1)->ll_namelen)
2519 return ERR_PTR(-ENAMETOOLONG);
2521 if (!lu_name_is_valid_2(name, namelen))
2522 return ERR_PTR(-EINVAL);
2525 if (op_data == NULL)
2526 OBD_ALLOC_PTR(op_data);
2528 if (op_data == NULL)
2529 return ERR_PTR(-ENOMEM);
2531 ll_i2gids(op_data->op_suppgids, i1, i2);
2532 op_data->op_fid1 = *ll_inode2fid(i1);
2533 op_data->op_capa1 = ll_mdscapa_get(i1);
2534 if (S_ISDIR(i1->i_mode))
2535 op_data->op_mea1 = ll_i2info(i1)->lli_lsm_md;
2538 op_data->op_fid2 = *ll_inode2fid(i2);
2539 op_data->op_capa2 = ll_mdscapa_get(i2);
2540 if (S_ISDIR(i2->i_mode))
2541 op_data->op_mea2 = ll_i2info(i2)->lli_lsm_md;
2543 fid_zero(&op_data->op_fid2);
2544 op_data->op_capa2 = NULL;
2547 if (ll_i2sbi(i1)->ll_flags & LL_SBI_64BIT_HASH)
2548 op_data->op_cli_flags |= CLI_HASH64;
2550 if (ll_need_32bit_api(ll_i2sbi(i1)))
2551 op_data->op_cli_flags |= CLI_API32;
2553 op_data->op_name = name;
2554 op_data->op_namelen = namelen;
2555 op_data->op_mode = mode;
2556 op_data->op_mod_time = cfs_time_current_sec();
2557 op_data->op_fsuid = from_kuid(&init_user_ns, current_fsuid());
2558 op_data->op_fsgid = from_kgid(&init_user_ns, current_fsgid());
2559 op_data->op_cap = cfs_curproc_cap_pack();
2560 op_data->op_bias = 0;
2561 op_data->op_cli_flags = 0;
2562 if ((opc == LUSTRE_OPC_CREATE) && (name != NULL) &&
2563 filename_is_volatile(name, namelen, &op_data->op_mds)) {
2564 op_data->op_bias |= MDS_CREATE_VOLATILE;
2566 op_data->op_mds = 0;
2568 op_data->op_data = data;
2570 /* When called by ll_setattr_raw, file is i1. */
2571 if (LLIF_DATA_MODIFIED & ll_i2info(i1)->lli_flags)
2572 op_data->op_bias |= MDS_DATA_MODIFIED;
2577 void ll_finish_md_op_data(struct md_op_data *op_data)
2579 capa_put(op_data->op_capa1);
2580 capa_put(op_data->op_capa2);
2581 OBD_FREE_PTR(op_data);
2584 #ifdef HAVE_SUPEROPS_USE_DENTRY
2585 int ll_show_options(struct seq_file *seq, struct dentry *dentry)
2587 int ll_show_options(struct seq_file *seq, struct vfsmount *vfs)
2590 struct ll_sb_info *sbi;
2592 #ifdef HAVE_SUPEROPS_USE_DENTRY
2593 LASSERT((seq != NULL) && (dentry != NULL));
2594 sbi = ll_s2sbi(dentry->d_sb);
2596 LASSERT((seq != NULL) && (vfs != NULL));
2597 sbi = ll_s2sbi(vfs->mnt_sb);
2600 if (sbi->ll_flags & LL_SBI_NOLCK)
2601 seq_puts(seq, ",nolock");
2603 if (sbi->ll_flags & LL_SBI_FLOCK)
2604 seq_puts(seq, ",flock");
2606 if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
2607 seq_puts(seq, ",localflock");
2609 if (sbi->ll_flags & LL_SBI_USER_XATTR)
2610 seq_puts(seq, ",user_xattr");
2612 if (sbi->ll_flags & LL_SBI_LAZYSTATFS)
2613 seq_puts(seq, ",lazystatfs");
2615 if (sbi->ll_flags & LL_SBI_USER_FID2PATH)
2616 seq_puts(seq, ",user_fid2path");
2622 * Get obd name by cmd, and copy out to user space
2624 int ll_get_obd_name(struct inode *inode, unsigned int cmd, unsigned long arg)
2626 struct ll_sb_info *sbi = ll_i2sbi(inode);
2627 struct obd_device *obd;
2630 if (cmd == OBD_IOC_GETDTNAME)
2631 obd = class_exp2obd(sbi->ll_dt_exp);
2632 else if (cmd == OBD_IOC_GETMDNAME)
2633 obd = class_exp2obd(sbi->ll_md_exp);
2640 if (copy_to_user((void __user *)arg, obd->obd_name,
2641 strlen(obd->obd_name) + 1))
2648 * Get lustre file system name by \a sbi. If \a buf is provided(non-NULL), the
2649 * fsname will be returned in this buffer; otherwise, a static buffer will be
2650 * used to store the fsname and returned to caller.
2652 char *ll_get_fsname(struct super_block *sb, char *buf, int buflen)
2654 static char fsname_static[MTI_NAME_MAXLEN];
2655 struct lustre_sb_info *lsi = s2lsi(sb);
2660 /* this means the caller wants to use static buffer
2661 * and it doesn't care about race. Usually this is
2662 * in error reporting path */
2663 buf = fsname_static;
2664 buflen = sizeof(fsname_static);
2667 len = strlen(lsi->lsi_lmd->lmd_profile);
2668 ptr = strrchr(lsi->lsi_lmd->lmd_profile, '-');
2669 if (ptr && (strcmp(ptr, "-client") == 0))
2672 if (unlikely(len >= buflen))
2674 strncpy(buf, lsi->lsi_lmd->lmd_profile, len);
2680 static char* ll_d_path(struct dentry *dentry, char *buf, int bufsize)
2687 p.mnt = current->fs->root.mnt;
2689 path = d_path(&p, buf, bufsize);
2694 void ll_dirty_page_discard_warn(struct page *page, int ioret)
2696 char *buf, *path = NULL;
2697 struct dentry *dentry = NULL;
2698 struct vvp_object *obj = cl_inode2vvp(page->mapping->host);
2700 /* this can be called inside spin lock so use GFP_ATOMIC. */
2701 buf = (char *)__get_free_page(GFP_ATOMIC);
2703 dentry = d_find_alias(page->mapping->host);
2705 path = ll_d_path(dentry, buf, PAGE_SIZE);
2709 "%s: dirty page discard: %s/fid: "DFID"/%s may get corrupted "
2710 "(rc %d)\n", ll_get_fsname(page->mapping->host->i_sb, NULL, 0),
2711 s2lsi(page->mapping->host->i_sb)->lsi_lmd->lmd_dev,
2712 PFID(&obj->vob_header.coh_lu.loh_fid),
2713 (path && !IS_ERR(path)) ? path : "", ioret);
2719 free_page((unsigned long)buf);
2722 ssize_t ll_copy_user_md(const struct lov_user_md __user *md,
2723 struct lov_user_md **kbuf)
2725 struct lov_user_md lum;
2729 if (copy_from_user(&lum, md, sizeof(lum)))
2732 lum_size = ll_lov_user_md_size(&lum);
2736 OBD_ALLOC(*kbuf, lum_size);
2740 if (copy_from_user(*kbuf, md, lum_size) != 0) {
2741 OBD_FREE(*kbuf, lum_size);
2749 * Compute llite root squash state after a change of root squash
2750 * configuration setting or add/remove of a lnet nid
2752 void ll_compute_rootsquash_state(struct ll_sb_info *sbi)
2754 struct root_squash_info *squash = &sbi->ll_squash;
2757 lnet_process_id_t id;
2759 /* Update norootsquash flag */
2760 down_write(&squash->rsi_sem);
2761 if (list_empty(&squash->rsi_nosquash_nids))
2762 sbi->ll_flags &= ~LL_SBI_NOROOTSQUASH;
2764 /* Do not apply root squash as soon as one of our NIDs is
2765 * in the nosquash_nids list */
2768 while (LNetGetId(i++, &id) != -ENOENT) {
2769 if (LNET_NETTYP(LNET_NIDNET(id.nid)) == LOLND)
2771 if (cfs_match_nid(id.nid, &squash->rsi_nosquash_nids)) {
2777 sbi->ll_flags |= LL_SBI_NOROOTSQUASH;
2779 sbi->ll_flags &= ~LL_SBI_NOROOTSQUASH;
2781 up_write(&squash->rsi_sem);
2785 * Parse linkea content to extract information about a given hardlink
2787 * \param[in] ldata - Initialized linkea data
2788 * \param[in] linkno - Link identifier
2789 * \param[out] parent_fid - The entry's parent FID
2790 * \param[out] ln - Entry name destination buffer
2792 * \retval 0 on success
2793 * \retval Appropriate negative error code on failure
2795 static int ll_linkea_decode(struct linkea_data *ldata, unsigned int linkno,
2796 struct lu_fid *parent_fid, struct lu_name *ln)
2802 rc = linkea_init(ldata);
2806 if (linkno >= ldata->ld_leh->leh_reccount)
2807 /* beyond last link */
2810 linkea_first_entry(ldata);
2811 for (idx = 0; ldata->ld_lee != NULL; idx++) {
2812 linkea_entry_unpack(ldata->ld_lee, &ldata->ld_reclen, ln,
2817 linkea_next_entry(ldata);
2827 * Get parent FID and name of an identified link. Operation is performed for
2828 * a given link number, letting the caller iterate over linkno to list one or
2829 * all links of an entry.
2831 * \param[in] file - File descriptor against which to perform the operation
2832 * \param[in,out] arg - User-filled structure containing the linkno to operate
2833 * on and the available size. It is eventually filled with
2834 * the requested information or left untouched on error
2836 * \retval - 0 on success
2837 * \retval - Appropriate negative error code on failure
2839 int ll_getparent(struct file *file, struct getparent __user *arg)
2841 struct dentry *dentry = file->f_dentry;
2842 struct inode *inode = file->f_dentry->d_inode;
2843 struct linkea_data *ldata;
2844 struct lu_buf buf = LU_BUF_NULL;
2846 struct lu_fid parent_fid;
2853 if (!cfs_capable(CFS_CAP_DAC_READ_SEARCH) &&
2854 !(ll_i2sbi(inode)->ll_flags & LL_SBI_USER_FID2PATH))
2857 if (get_user(name_size, &arg->gp_name_size))
2860 if (get_user(linkno, &arg->gp_linkno))
2863 if (name_size > PATH_MAX)
2866 OBD_ALLOC(ldata, sizeof(*ldata));
2870 rc = linkea_data_new(ldata, &buf);
2872 GOTO(ldata_free, rc);
2874 rc = ll_getxattr(dentry, XATTR_NAME_LINK, buf.lb_buf, buf.lb_len);
2878 rc = ll_linkea_decode(ldata, linkno, &parent_fid, &ln);
2882 if (ln.ln_namelen >= name_size)
2883 GOTO(lb_free, rc = -EOVERFLOW);
2885 if (copy_to_user(&arg->gp_fid, &parent_fid, sizeof(arg->gp_fid)))
2886 GOTO(lb_free, rc = -EFAULT);
2888 if (copy_to_user(&arg->gp_name, ln.ln_name, ln.ln_namelen))
2889 GOTO(lb_free, rc = -EFAULT);
2891 if (put_user('\0', arg->gp_name + ln.ln_namelen))
2892 GOTO(lb_free, rc = -EFAULT);
2897 OBD_FREE(ldata, sizeof(*ldata));