4 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
6 * This program is free software; you can redistribute it and/or modify
7 * it under the terms of the GNU General Public License version 2 only,
8 * as published by the Free Software Foundation.
10 * This program is distributed in the hope that it will be useful, but
11 * WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
13 * General Public License version 2 for more details (a copy is included
14 * in the LICENSE file that accompanied this code).
16 * You should have received a copy of the GNU General Public License
17 * version 2 along with this program; If not, see
18 * http://www.gnu.org/licenses/gpl-2.0.html
23 * Copyright (c) 2003, 2010, Oracle and/or its affiliates. All rights reserved.
24 * Use is subject to license terms.
26 * Copyright (c) 2011, 2017, Intel Corporation.
29 * This file is part of Lustre, http://www.lustre.org/
30 * Lustre is a trademark of Sun Microsystems, Inc.
32 * lustre/llite/llite_lib.c
34 * Lustre Light Super operations
37 #define DEBUG_SUBSYSTEM S_LLITE
39 #include <linux/cpu.h>
40 #include <linux/module.h>
41 #include <linux/random.h>
42 #include <linux/statfs.h>
43 #include <linux/time.h>
44 #include <linux/types.h>
45 #include <libcfs/linux/linux-uuid.h>
46 #include <linux/version.h>
48 #include <linux/user_namespace.h>
49 #include <linux/delay.h>
50 #include <linux/uidgid.h>
51 #include <linux/security.h>
52 #include <linux/fs_struct.h>
54 #ifndef HAVE_CPUS_READ_LOCK
55 #include <libcfs/linux/linux-cpu.h>
57 #include <uapi/linux/lustre/lustre_ioctl.h>
58 #ifdef HAVE_UAPI_LINUX_MOUNT_H
59 #include <uapi/linux/mount.h>
62 #include <lustre_ha.h>
63 #include <lustre_dlm.h>
64 #include <lprocfs_status.h>
65 #include <lustre_disk.h>
66 #include <uapi/linux/lustre/lustre_param.h>
67 #include <lustre_log.h>
68 #include <cl_object.h>
69 #include <obd_cksum.h>
70 #include "llite_internal.h"
72 struct kmem_cache *ll_file_data_slab;
75 #define log2(n) ffz(~(n))
79 * If there is only one number of core visible to Lustre,
80 * async readahead will be disabled, to avoid massive over
81 * subscription, we use 1/2 of active cores as default max
82 * async readahead requests.
84 static inline unsigned int ll_get_ra_async_max_active(void)
86 return cfs_cpt_weight(cfs_cpt_tab, CFS_CPT_ANY) >> 1;
89 static struct ll_sb_info *ll_init_sbi(void)
91 struct ll_sb_info *sbi = NULL;
93 unsigned long lru_page_max;
102 RETURN(ERR_PTR(-ENOMEM));
104 rc = pcc_super_init(&sbi->ll_pcc_super);
108 spin_lock_init(&sbi->ll_lock);
109 mutex_init(&sbi->ll_lco.lco_lock);
110 spin_lock_init(&sbi->ll_pp_extent_lock);
111 spin_lock_init(&sbi->ll_process_lock);
112 sbi->ll_rw_stats_on = 0;
113 sbi->ll_statfs_max_age = OBD_STATFS_CACHE_SECONDS;
116 pages = si.totalram - si.totalhigh;
117 lru_page_max = pages / 2;
119 sbi->ll_ra_info.ra_async_max_active = ll_get_ra_async_max_active();
120 sbi->ll_ra_info.ll_readahead_wq =
121 cfs_cpt_bind_workqueue("ll-readahead-wq", cfs_cpt_tab,
123 sbi->ll_ra_info.ra_async_max_active);
124 if (IS_ERR(sbi->ll_ra_info.ll_readahead_wq))
125 GOTO(out_pcc, rc = PTR_ERR(sbi->ll_ra_info.ll_readahead_wq));
127 /* initialize ll_cache data */
128 sbi->ll_cache = cl_cache_init(lru_page_max);
129 if (sbi->ll_cache == NULL)
130 GOTO(out_destroy_ra, rc = -ENOMEM);
132 sbi->ll_ra_info.ra_max_pages_per_file = min(pages / 32,
133 SBI_DEFAULT_READ_AHEAD_MAX);
134 sbi->ll_ra_info.ra_async_pages_per_file_threshold =
135 sbi->ll_ra_info.ra_max_pages_per_file;
136 sbi->ll_ra_info.ra_max_pages = sbi->ll_ra_info.ra_max_pages_per_file;
137 sbi->ll_ra_info.ra_max_read_ahead_whole_pages = -1;
138 atomic_set(&sbi->ll_ra_info.ra_async_inflight, 0);
140 sbi->ll_flags |= LL_SBI_VERBOSE;
141 #ifdef ENABLE_CHECKSUM
142 sbi->ll_flags |= LL_SBI_CHECKSUM;
145 sbi->ll_flags |= LL_SBI_FLOCK;
148 #ifdef HAVE_LRU_RESIZE_SUPPORT
149 sbi->ll_flags |= LL_SBI_LRU_RESIZE;
151 sbi->ll_flags |= LL_SBI_LAZYSTATFS;
153 for (i = 0; i <= LL_PROCESS_HIST_MAX; i++) {
154 spin_lock_init(&sbi->ll_rw_extents_info.pp_extents[i].
156 spin_lock_init(&sbi->ll_rw_extents_info.pp_extents[i].
160 /* metadata statahead is enabled by default */
161 sbi->ll_sa_running_max = LL_SA_RUNNING_DEF;
162 sbi->ll_sa_max = LL_SA_RPC_DEF;
163 atomic_set(&sbi->ll_sa_total, 0);
164 atomic_set(&sbi->ll_sa_wrong, 0);
165 atomic_set(&sbi->ll_sa_running, 0);
166 atomic_set(&sbi->ll_agl_total, 0);
167 sbi->ll_flags |= LL_SBI_AGL_ENABLED;
168 sbi->ll_flags |= LL_SBI_FAST_READ;
169 sbi->ll_flags |= LL_SBI_TINY_WRITE;
170 ll_sbi_set_encrypt(sbi, true);
173 sbi->ll_squash.rsi_uid = 0;
174 sbi->ll_squash.rsi_gid = 0;
175 INIT_LIST_HEAD(&sbi->ll_squash.rsi_nosquash_nids);
176 spin_lock_init(&sbi->ll_squash.rsi_lock);
178 /* Per-filesystem file heat */
179 sbi->ll_heat_decay_weight = SBI_DEFAULT_HEAT_DECAY_WEIGHT;
180 sbi->ll_heat_period_second = SBI_DEFAULT_HEAT_PERIOD_SECOND;
183 destroy_workqueue(sbi->ll_ra_info.ll_readahead_wq);
185 pcc_super_fini(&sbi->ll_pcc_super);
191 static void ll_free_sbi(struct super_block *sb)
193 struct ll_sb_info *sbi = ll_s2sbi(sb);
197 if (!list_empty(&sbi->ll_squash.rsi_nosquash_nids))
198 cfs_free_nidlist(&sbi->ll_squash.rsi_nosquash_nids);
199 if (sbi->ll_ra_info.ll_readahead_wq)
200 destroy_workqueue(sbi->ll_ra_info.ll_readahead_wq);
201 if (sbi->ll_cache != NULL) {
202 cl_cache_decref(sbi->ll_cache);
203 sbi->ll_cache = NULL;
205 pcc_super_fini(&sbi->ll_pcc_super);
206 OBD_FREE(sbi, sizeof(*sbi));
211 static int client_common_fill_super(struct super_block *sb, char *md, char *dt)
213 struct inode *root = NULL;
214 struct ll_sb_info *sbi = ll_s2sbi(sb);
215 struct obd_statfs *osfs = NULL;
216 struct ptlrpc_request *request = NULL;
217 struct obd_connect_data *data = NULL;
218 struct obd_uuid *uuid;
219 struct md_op_data *op_data;
220 struct lustre_md lmd;
222 int size, err, checksum;
225 sbi->ll_md_obd = class_name2obd(md);
226 if (!sbi->ll_md_obd) {
227 CERROR("MD %s: not setup or attached\n", md);
241 /* pass client page size via ocd_grant_blkbits, the server should report
242 * back its backend blocksize for grant calculation purpose */
243 data->ocd_grant_blkbits = PAGE_SHIFT;
245 /* indicate MDT features supported by this client */
246 data->ocd_connect_flags = OBD_CONNECT_IBITS | OBD_CONNECT_NODEVOH |
247 OBD_CONNECT_ATTRFID | OBD_CONNECT_GRANT |
248 OBD_CONNECT_VERSION | OBD_CONNECT_BRW_SIZE |
249 OBD_CONNECT_SRVLOCK | OBD_CONNECT_TRUNCLOCK|
250 OBD_CONNECT_MDS_CAPA | OBD_CONNECT_OSS_CAPA |
251 OBD_CONNECT_CANCELSET | OBD_CONNECT_FID |
252 OBD_CONNECT_AT | OBD_CONNECT_LOV_V3 |
253 OBD_CONNECT_VBR | OBD_CONNECT_FULL20 |
254 OBD_CONNECT_64BITHASH |
255 OBD_CONNECT_EINPROGRESS |
256 OBD_CONNECT_JOBSTATS | OBD_CONNECT_LVB_TYPE |
257 OBD_CONNECT_LAYOUTLOCK | OBD_CONNECT_PINGLESS|
258 OBD_CONNECT_MAX_EASIZE |
259 OBD_CONNECT_FLOCK_DEAD |
260 OBD_CONNECT_DISP_STRIPE | OBD_CONNECT_LFSCK |
261 OBD_CONNECT_OPEN_BY_FID |
262 OBD_CONNECT_DIR_STRIPE |
263 OBD_CONNECT_BULK_MBITS | OBD_CONNECT_CKSUM |
264 OBD_CONNECT_SUBTREE |
265 OBD_CONNECT_MULTIMODRPCS |
266 OBD_CONNECT_GRANT_PARAM |
267 OBD_CONNECT_SHORTIO | OBD_CONNECT_FLAGS2;
269 data->ocd_connect_flags2 = OBD_CONNECT2_DIR_MIGRATE |
270 OBD_CONNECT2_SUM_STATFS |
271 OBD_CONNECT2_OVERSTRIPING |
273 OBD_CONNECT2_LOCK_CONVERT |
274 OBD_CONNECT2_ARCHIVE_ID_ARRAY |
275 OBD_CONNECT2_INC_XID |
277 OBD_CONNECT2_ASYNC_DISCARD |
281 #ifdef HAVE_LRU_RESIZE_SUPPORT
282 if (sbi->ll_flags & LL_SBI_LRU_RESIZE)
283 data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
285 data->ocd_connect_flags |= OBD_CONNECT_ACL_FLAGS;
287 data->ocd_cksum_types = obd_cksum_types_supported_client();
289 if (OBD_FAIL_CHECK(OBD_FAIL_MDC_LIGHTWEIGHT))
290 /* flag mdc connection as lightweight, only used for test
291 * purpose, use with care */
292 data->ocd_connect_flags |= OBD_CONNECT_LIGHTWEIGHT;
294 data->ocd_ibits_known = MDS_INODELOCK_FULL;
295 data->ocd_version = LUSTRE_VERSION_CODE;
297 if (sb->s_flags & SB_RDONLY)
298 data->ocd_connect_flags |= OBD_CONNECT_RDONLY;
299 if (sbi->ll_flags & LL_SBI_USER_XATTR)
300 data->ocd_connect_flags |= OBD_CONNECT_XATTR;
303 /* Setting this indicates we correctly support S_NOSEC (See kernel
304 * commit 9e1f1de02c2275d7172e18dc4e7c2065777611bf)
306 sb->s_flags |= SB_NOSEC;
309 if (sbi->ll_flags & LL_SBI_FLOCK)
310 sbi->ll_fop = &ll_file_operations_flock;
311 else if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
312 sbi->ll_fop = &ll_file_operations;
314 sbi->ll_fop = &ll_file_operations_noflock;
316 /* always ping even if server suppress_pings */
317 if (sbi->ll_flags & LL_SBI_ALWAYS_PING)
318 data->ocd_connect_flags &= ~OBD_CONNECT_PINGLESS;
320 obd_connect_set_secctx(data);
321 if (ll_sbi_has_encrypt(sbi))
322 obd_connect_set_enc(data);
324 #if defined(CONFIG_SECURITY)
325 data->ocd_connect_flags2 |= OBD_CONNECT2_SELINUX_POLICY;
328 data->ocd_brw_size = MD_MAX_BRW_SIZE;
330 err = obd_connect(NULL, &sbi->ll_md_exp, sbi->ll_md_obd,
331 &sbi->ll_sb_uuid, data, sbi->ll_cache);
333 LCONSOLE_ERROR_MSG(0x14f, "An MDT (md %s) is performing "
334 "recovery, of which this client is not a "
335 "part. Please wait for recovery to complete,"
336 " abort, or time out.\n", md);
339 CERROR("cannot connect to %s: rc = %d\n", md, err);
343 sbi->ll_md_exp->exp_connect_data = *data;
345 err = obd_fid_init(sbi->ll_md_exp->exp_obd, sbi->ll_md_exp,
346 LUSTRE_SEQ_METADATA);
348 CERROR("%s: Can't init metadata layer FID infrastructure, "
349 "rc = %d\n", sbi->ll_md_exp->exp_obd->obd_name, err);
353 /* For mount, we only need fs info from MDT0, and also in DNE, it
354 * can make sure the client can be mounted as long as MDT0 is
356 err = obd_statfs(NULL, sbi->ll_md_exp, osfs,
357 ktime_get_seconds() - sbi->ll_statfs_max_age,
358 OBD_STATFS_FOR_MDT0);
360 GOTO(out_md_fid, err);
362 /* This needs to be after statfs to ensure connect has finished.
363 * Note that "data" does NOT contain the valid connect reply.
364 * If connecting to a 1.8 server there will be no LMV device, so
365 * we can access the MDC export directly and exp_connect_flags will
366 * be non-zero, but if accessing an upgraded 2.1 server it will
367 * have the correct flags filled in.
368 * XXX: fill in the LMV exp_connect_flags from MDC(s). */
369 valid = exp_connect_flags(sbi->ll_md_exp) & CLIENT_CONNECT_MDT_REQD;
370 if (exp_connect_flags(sbi->ll_md_exp) != 0 &&
371 valid != CLIENT_CONNECT_MDT_REQD) {
374 OBD_ALLOC_WAIT(buf, PAGE_SIZE);
375 obd_connect_flags2str(buf, PAGE_SIZE,
376 valid ^ CLIENT_CONNECT_MDT_REQD, 0, ",");
377 LCONSOLE_ERROR_MSG(0x170, "Server %s does not support "
378 "feature(s) needed for correct operation "
379 "of this client (%s). Please upgrade "
380 "server or downgrade client.\n",
381 sbi->ll_md_exp->exp_obd->obd_name, buf);
382 OBD_FREE(buf, PAGE_SIZE);
383 GOTO(out_md_fid, err = -EPROTO);
386 size = sizeof(*data);
387 err = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_CONN_DATA),
388 KEY_CONN_DATA, &size, data);
390 CERROR("%s: Get connect data failed: rc = %d\n",
391 sbi->ll_md_exp->exp_obd->obd_name, err);
392 GOTO(out_md_fid, err);
395 LASSERT(osfs->os_bsize);
396 sb->s_blocksize = osfs->os_bsize;
397 sb->s_blocksize_bits = log2(osfs->os_bsize);
398 sb->s_magic = LL_SUPER_MAGIC;
399 sb->s_maxbytes = MAX_LFS_FILESIZE;
400 sbi->ll_namelen = osfs->os_namelen;
401 sbi->ll_mnt.mnt = current->fs->root.mnt;
403 if ((sbi->ll_flags & LL_SBI_USER_XATTR) &&
404 !(data->ocd_connect_flags & OBD_CONNECT_XATTR)) {
405 LCONSOLE_INFO("Disabling user_xattr feature because "
406 "it is not supported on the server\n");
407 sbi->ll_flags &= ~LL_SBI_USER_XATTR;
410 if (data->ocd_connect_flags & OBD_CONNECT_ACL) {
412 sb->s_flags |= SB_POSIXACL;
414 sbi->ll_flags |= LL_SBI_ACL;
416 LCONSOLE_INFO("client wants to enable acl, but mdt not!\n");
418 sb->s_flags &= ~SB_POSIXACL;
420 sbi->ll_flags &= ~LL_SBI_ACL;
423 if (data->ocd_connect_flags & OBD_CONNECT_64BITHASH)
424 sbi->ll_flags |= LL_SBI_64BIT_HASH;
426 if (data->ocd_connect_flags & OBD_CONNECT_LAYOUTLOCK)
427 sbi->ll_flags |= LL_SBI_LAYOUT_LOCK;
429 if (obd_connect_has_secctx(data))
430 sbi->ll_flags |= LL_SBI_FILE_SECCTX;
432 if (ll_sbi_has_encrypt(sbi) && !obd_connect_has_enc(data)) {
433 if (ll_sbi_has_test_dummy_encryption(sbi))
434 LCONSOLE_WARN("%s: server %s does not support encryption feature, encryption deactivated.\n",
436 sbi->ll_md_exp->exp_obd->obd_name);
437 ll_sbi_set_encrypt(sbi, false);
440 if (data->ocd_ibits_known & MDS_INODELOCK_XATTR) {
441 if (!(data->ocd_connect_flags & OBD_CONNECT_MAX_EASIZE)) {
442 LCONSOLE_INFO("%s: disabling xattr cache due to "
443 "unknown maximum xattr size.\n", dt);
444 } else if (!sbi->ll_xattr_cache_set) {
445 /* If xattr_cache is already set (no matter 0 or 1)
446 * during processing llog, it won't be enabled here. */
447 sbi->ll_flags |= LL_SBI_XATTR_CACHE;
448 sbi->ll_xattr_cache_enabled = 1;
452 sbi->ll_dt_obd = class_name2obd(dt);
453 if (!sbi->ll_dt_obd) {
454 CERROR("DT %s: not setup or attached\n", dt);
455 GOTO(out_md_fid, err = -ENODEV);
458 /* pass client page size via ocd_grant_blkbits, the server should report
459 * back its backend blocksize for grant calculation purpose */
460 data->ocd_grant_blkbits = PAGE_SHIFT;
462 /* indicate OST features supported by this client */
463 data->ocd_connect_flags = OBD_CONNECT_GRANT | OBD_CONNECT_VERSION |
464 OBD_CONNECT_REQPORTAL | OBD_CONNECT_BRW_SIZE |
465 OBD_CONNECT_CANCELSET | OBD_CONNECT_FID |
466 OBD_CONNECT_SRVLOCK | OBD_CONNECT_TRUNCLOCK|
467 OBD_CONNECT_AT | OBD_CONNECT_OSS_CAPA |
468 OBD_CONNECT_VBR | OBD_CONNECT_FULL20 |
469 OBD_CONNECT_64BITHASH | OBD_CONNECT_MAXBYTES |
470 OBD_CONNECT_EINPROGRESS |
471 OBD_CONNECT_JOBSTATS | OBD_CONNECT_LVB_TYPE |
472 OBD_CONNECT_LAYOUTLOCK |
473 OBD_CONNECT_PINGLESS | OBD_CONNECT_LFSCK |
474 OBD_CONNECT_BULK_MBITS | OBD_CONNECT_SHORTIO |
475 OBD_CONNECT_FLAGS2 | OBD_CONNECT_GRANT_SHRINK;
476 data->ocd_connect_flags2 = OBD_CONNECT2_LOCKAHEAD |
477 OBD_CONNECT2_INC_XID;
479 if (!OBD_FAIL_CHECK(OBD_FAIL_OSC_CONNECT_GRANT_PARAM))
480 data->ocd_connect_flags |= OBD_CONNECT_GRANT_PARAM;
482 /* OBD_CONNECT_CKSUM should always be set, even if checksums are
483 * disabled by default, because it can still be enabled on the
484 * fly via /sys. As a consequence, we still need to come to an
485 * agreement on the supported algorithms at connect time
487 data->ocd_connect_flags |= OBD_CONNECT_CKSUM;
489 if (OBD_FAIL_CHECK(OBD_FAIL_OSC_CKSUM_ADLER_ONLY))
490 data->ocd_cksum_types = OBD_CKSUM_ADLER;
492 data->ocd_cksum_types = obd_cksum_types_supported_client();
494 #ifdef HAVE_LRU_RESIZE_SUPPORT
495 data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
497 /* always ping even if server suppress_pings */
498 if (sbi->ll_flags & LL_SBI_ALWAYS_PING)
499 data->ocd_connect_flags &= ~OBD_CONNECT_PINGLESS;
501 if (ll_sbi_has_encrypt(sbi))
502 obd_connect_set_enc(data);
504 CDEBUG(D_RPCTRACE, "ocd_connect_flags: %#llx ocd_version: %d "
505 "ocd_grant: %d\n", data->ocd_connect_flags,
506 data->ocd_version, data->ocd_grant);
508 sbi->ll_dt_obd->obd_upcall.onu_owner = &sbi->ll_lco;
509 sbi->ll_dt_obd->obd_upcall.onu_upcall = cl_ocd_update;
511 data->ocd_brw_size = DT_MAX_BRW_SIZE;
513 err = obd_connect(NULL, &sbi->ll_dt_exp, sbi->ll_dt_obd,
514 &sbi->ll_sb_uuid, data, sbi->ll_cache);
516 LCONSOLE_ERROR_MSG(0x150, "An OST (dt %s) is performing "
517 "recovery, of which this client is not a "
518 "part. Please wait for recovery to "
519 "complete, abort, or time out.\n", dt);
522 CERROR("%s: Cannot connect to %s: rc = %d\n",
523 sbi->ll_dt_exp->exp_obd->obd_name, dt, err);
527 if (ll_sbi_has_encrypt(sbi) &&
528 !obd_connect_has_enc(&sbi->ll_dt_obd->u.lov.lov_ocd)) {
529 if (ll_sbi_has_test_dummy_encryption(sbi))
530 LCONSOLE_WARN("%s: server %s does not support encryption feature, encryption deactivated.\n",
532 ll_sbi_set_encrypt(sbi, false);
533 } else if (ll_sbi_has_test_dummy_encryption(sbi)) {
534 LCONSOLE_WARN("Test dummy encryption mode enabled\n");
537 sbi->ll_dt_exp->exp_connect_data = *data;
539 /* Don't change value if it was specified in the config log */
540 if (sbi->ll_ra_info.ra_max_read_ahead_whole_pages == -1) {
541 sbi->ll_ra_info.ra_max_read_ahead_whole_pages =
542 max_t(unsigned long, SBI_DEFAULT_READ_AHEAD_WHOLE_MAX,
543 (data->ocd_brw_size >> PAGE_SHIFT));
544 if (sbi->ll_ra_info.ra_max_read_ahead_whole_pages >
545 sbi->ll_ra_info.ra_max_pages_per_file)
546 sbi->ll_ra_info.ra_max_read_ahead_whole_pages =
547 sbi->ll_ra_info.ra_max_pages_per_file;
550 err = obd_fid_init(sbi->ll_dt_exp->exp_obd, sbi->ll_dt_exp,
551 LUSTRE_SEQ_METADATA);
553 CERROR("%s: Can't init data layer FID infrastructure, "
554 "rc = %d\n", sbi->ll_dt_exp->exp_obd->obd_name, err);
558 mutex_lock(&sbi->ll_lco.lco_lock);
559 sbi->ll_lco.lco_flags = data->ocd_connect_flags;
560 sbi->ll_lco.lco_md_exp = sbi->ll_md_exp;
561 sbi->ll_lco.lco_dt_exp = sbi->ll_dt_exp;
562 mutex_unlock(&sbi->ll_lco.lco_lock);
564 fid_zero(&sbi->ll_root_fid);
565 err = md_get_root(sbi->ll_md_exp, get_mount_fileset(sb),
568 CERROR("cannot mds_connect: rc = %d\n", err);
569 GOTO(out_lock_cn_cb, err);
571 if (!fid_is_sane(&sbi->ll_root_fid)) {
572 CERROR("%s: Invalid root fid "DFID" during mount\n",
573 sbi->ll_md_exp->exp_obd->obd_name,
574 PFID(&sbi->ll_root_fid));
575 GOTO(out_lock_cn_cb, err = -EINVAL);
577 CDEBUG(D_SUPER, "rootfid "DFID"\n", PFID(&sbi->ll_root_fid));
579 sb->s_op = &lustre_super_operations;
580 sb->s_xattr = ll_xattr_handlers;
581 #if THREAD_SIZE >= 8192 /*b=17630*/
582 sb->s_export_op = &lustre_export_operations;
584 #ifdef HAVE_LUSTRE_CRYPTO
585 llcrypt_set_ops(sb, &lustre_cryptops);
589 * XXX: move this to after cbd setup? */
590 valid = OBD_MD_FLGETATTR | OBD_MD_FLBLOCKS | OBD_MD_FLMODEASIZE;
591 if (sbi->ll_flags & LL_SBI_ACL)
592 valid |= OBD_MD_FLACL;
594 OBD_ALLOC_PTR(op_data);
596 GOTO(out_lock_cn_cb, err = -ENOMEM);
598 op_data->op_fid1 = sbi->ll_root_fid;
599 op_data->op_mode = 0;
600 op_data->op_valid = valid;
602 err = md_getattr(sbi->ll_md_exp, op_data, &request);
604 OBD_FREE_PTR(op_data);
606 CERROR("%s: md_getattr failed for root: rc = %d\n",
607 sbi->ll_md_exp->exp_obd->obd_name, err);
608 GOTO(out_lock_cn_cb, err);
611 err = md_get_lustre_md(sbi->ll_md_exp, request, sbi->ll_dt_exp,
612 sbi->ll_md_exp, &lmd);
614 CERROR("failed to understand root inode md: rc = %d\n", err);
615 ptlrpc_req_finished(request);
616 GOTO(out_lock_cn_cb, err);
619 LASSERT(fid_is_sane(&sbi->ll_root_fid));
620 root = ll_iget(sb, cl_fid_build_ino(&sbi->ll_root_fid,
621 sbi->ll_flags & LL_SBI_32BIT_API),
623 md_free_lustre_md(sbi->ll_md_exp, &lmd);
624 ptlrpc_req_finished(request);
628 err = IS_ERR(root) ? PTR_ERR(root) : -EBADF;
630 CERROR("%s: bad ll_iget() for root: rc = %d\n",
631 sbi->ll_fsname, err);
635 checksum = sbi->ll_flags & LL_SBI_CHECKSUM;
636 if (sbi->ll_checksum_set) {
637 err = obd_set_info_async(NULL, sbi->ll_dt_exp,
638 sizeof(KEY_CHECKSUM), KEY_CHECKSUM,
639 sizeof(checksum), &checksum, NULL);
641 CERROR("%s: Set checksum failed: rc = %d\n",
642 sbi->ll_dt_exp->exp_obd->obd_name, err);
648 sb->s_root = d_make_root(root);
649 if (sb->s_root == NULL) {
651 CERROR("%s: can't make root dentry: rc = %d\n",
652 sbi->ll_fsname, err);
656 sbi->ll_sdev_orig = sb->s_dev;
658 /* We set sb->s_dev equal on all lustre clients in order to support
659 * NFS export clustering. NFSD requires that the FSID be the same
661 /* s_dev is also used in lt_compare() to compare two fs, but that is
662 * only a node-local comparison. */
663 uuid = obd_get_uuid(sbi->ll_md_exp);
665 sb->s_dev = get_uuid2int(uuid->uuid, strlen(uuid->uuid));
672 if (sbi->ll_dt_obd) {
673 err = sysfs_create_link(&sbi->ll_kset.kobj,
674 &sbi->ll_dt_obd->obd_kset.kobj,
675 sbi->ll_dt_obd->obd_type->typ_name);
677 CERROR("%s: could not register %s in llite: rc = %d\n",
678 dt, sbi->ll_fsname, err);
683 if (sbi->ll_md_obd) {
684 err = sysfs_create_link(&sbi->ll_kset.kobj,
685 &sbi->ll_md_obd->obd_kset.kobj,
686 sbi->ll_md_obd->obd_type->typ_name);
688 CERROR("%s: could not register %s in llite: rc = %d\n",
689 md, sbi->ll_fsname, err);
699 obd_fid_fini(sbi->ll_dt_exp->exp_obd);
701 obd_disconnect(sbi->ll_dt_exp);
702 sbi->ll_dt_exp = NULL;
703 sbi->ll_dt_obd = NULL;
705 obd_fid_fini(sbi->ll_md_exp->exp_obd);
707 obd_disconnect(sbi->ll_md_exp);
708 sbi->ll_md_exp = NULL;
709 sbi->ll_md_obd = NULL;
718 int ll_get_max_mdsize(struct ll_sb_info *sbi, int *lmmsize)
722 size = sizeof(*lmmsize);
723 rc = obd_get_info(NULL, sbi->ll_dt_exp, sizeof(KEY_MAX_EASIZE),
724 KEY_MAX_EASIZE, &size, lmmsize);
726 CERROR("%s: cannot get max LOV EA size: rc = %d\n",
727 sbi->ll_dt_exp->exp_obd->obd_name, rc);
731 CDEBUG(D_INFO, "max LOV ea size: %d\n", *lmmsize);
734 rc = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_MAX_EASIZE),
735 KEY_MAX_EASIZE, &size, lmmsize);
737 CERROR("Get max mdsize error rc %d\n", rc);
739 CDEBUG(D_INFO, "max LMV ea size: %d\n", *lmmsize);
745 * Get the value of the default_easize parameter.
747 * \see client_obd::cl_default_mds_easize
749 * \param[in] sbi superblock info for this filesystem
750 * \param[out] lmmsize pointer to storage location for value
752 * \retval 0 on success
753 * \retval negative negated errno on failure
755 int ll_get_default_mdsize(struct ll_sb_info *sbi, int *lmmsize)
760 rc = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_DEFAULT_EASIZE),
761 KEY_DEFAULT_EASIZE, &size, lmmsize);
763 CERROR("Get default mdsize error rc %d\n", rc);
769 * Set the default_easize parameter to the given value.
771 * \see client_obd::cl_default_mds_easize
773 * \param[in] sbi superblock info for this filesystem
774 * \param[in] lmmsize the size to set
776 * \retval 0 on success
777 * \retval negative negated errno on failure
779 int ll_set_default_mdsize(struct ll_sb_info *sbi, int lmmsize)
783 if (lmmsize < sizeof(struct lov_mds_md) ||
784 lmmsize > OBD_MAX_DEFAULT_EA_SIZE)
787 rc = obd_set_info_async(NULL, sbi->ll_md_exp,
788 sizeof(KEY_DEFAULT_EASIZE), KEY_DEFAULT_EASIZE,
789 sizeof(int), &lmmsize, NULL);
794 static void client_common_put_super(struct super_block *sb)
796 struct ll_sb_info *sbi = ll_s2sbi(sb);
801 obd_fid_fini(sbi->ll_dt_exp->exp_obd);
802 obd_disconnect(sbi->ll_dt_exp);
803 sbi->ll_dt_exp = NULL;
805 ll_debugfs_unregister_super(sb);
807 obd_fid_fini(sbi->ll_md_exp->exp_obd);
808 obd_disconnect(sbi->ll_md_exp);
809 sbi->ll_md_exp = NULL;
814 void ll_kill_super(struct super_block *sb)
816 struct ll_sb_info *sbi;
820 if (!(sb->s_flags & SB_ACTIVE))
824 /* we need restore s_dev from changed for clustred NFS before put_super
825 * because new kernels have cached s_dev and change sb->s_dev in
826 * put_super not affected real removing devices */
828 sb->s_dev = sbi->ll_sdev_orig;
830 /* wait running statahead threads to quit */
831 while (atomic_read(&sbi->ll_sa_running) > 0)
832 schedule_timeout_uninterruptible(
833 cfs_time_seconds(1) >> 3);
839 static inline int ll_set_opt(const char *opt, char *data, int fl)
841 if (strncmp(opt, data, strlen(opt)) != 0)
847 /* non-client-specific mount options are parsed in lmd_parse */
848 static int ll_options(char *options, struct ll_sb_info *sbi)
851 char *s1 = options, *s2;
852 int *flags = &sbi->ll_flags;
858 CDEBUG(D_CONFIG, "Parsing opts %s\n", options);
861 CDEBUG(D_SUPER, "next opt=%s\n", s1);
862 tmp = ll_set_opt("nolock", s1, LL_SBI_NOLCK);
867 tmp = ll_set_opt("flock", s1, LL_SBI_FLOCK);
869 *flags = (*flags & ~LL_SBI_LOCALFLOCK) | tmp;
872 tmp = ll_set_opt("localflock", s1, LL_SBI_LOCALFLOCK);
874 *flags = (*flags & ~LL_SBI_FLOCK) | tmp;
877 tmp = ll_set_opt("noflock", s1, LL_SBI_FLOCK|LL_SBI_LOCALFLOCK);
882 tmp = ll_set_opt("user_xattr", s1, LL_SBI_USER_XATTR);
887 tmp = ll_set_opt("nouser_xattr", s1, LL_SBI_USER_XATTR);
892 tmp = ll_set_opt("context", s1, 1);
895 tmp = ll_set_opt("fscontext", s1, 1);
898 tmp = ll_set_opt("defcontext", s1, 1);
901 tmp = ll_set_opt("rootcontext", s1, 1);
904 tmp = ll_set_opt("user_fid2path", s1, LL_SBI_USER_FID2PATH);
909 tmp = ll_set_opt("nouser_fid2path", s1, LL_SBI_USER_FID2PATH);
915 tmp = ll_set_opt("checksum", s1, LL_SBI_CHECKSUM);
918 sbi->ll_checksum_set = 1;
921 tmp = ll_set_opt("nochecksum", s1, LL_SBI_CHECKSUM);
924 sbi->ll_checksum_set = 1;
927 tmp = ll_set_opt("lruresize", s1, LL_SBI_LRU_RESIZE);
932 tmp = ll_set_opt("nolruresize", s1, LL_SBI_LRU_RESIZE);
937 tmp = ll_set_opt("lazystatfs", s1, LL_SBI_LAZYSTATFS);
942 tmp = ll_set_opt("nolazystatfs", s1, LL_SBI_LAZYSTATFS);
947 tmp = ll_set_opt("32bitapi", s1, LL_SBI_32BIT_API);
952 tmp = ll_set_opt("verbose", s1, LL_SBI_VERBOSE);
957 tmp = ll_set_opt("noverbose", s1, LL_SBI_VERBOSE);
962 tmp = ll_set_opt("always_ping", s1, LL_SBI_ALWAYS_PING);
967 tmp = ll_set_opt("test_dummy_encryption", s1,
968 LL_SBI_TEST_DUMMY_ENCRYPTION);
970 #ifdef HAVE_LUSTRE_CRYPTO
973 LCONSOLE_WARN("Test dummy encryption mount option ignored: encryption not supported\n");
977 tmp = ll_set_opt("noencrypt", s1, LL_SBI_ENCRYPT);
979 #ifdef HAVE_LUSTRE_CRYPTO
982 LCONSOLE_WARN("noencrypt mount option ignored: encryption not supported\n");
986 LCONSOLE_ERROR_MSG(0x152, "Unknown option '%s', won't mount.\n",
992 s2 = strchr(s1, ',');
1000 void ll_lli_init(struct ll_inode_info *lli)
1002 lli->lli_inode_magic = LLI_INODE_MAGIC;
1004 spin_lock_init(&lli->lli_lock);
1005 lli->lli_posix_acl = NULL;
1006 /* Do not set lli_fid, it has been initialized already. */
1007 fid_zero(&lli->lli_pfid);
1008 lli->lli_mds_read_och = NULL;
1009 lli->lli_mds_write_och = NULL;
1010 lli->lli_mds_exec_och = NULL;
1011 lli->lli_open_fd_read_count = 0;
1012 lli->lli_open_fd_write_count = 0;
1013 lli->lli_open_fd_exec_count = 0;
1014 mutex_init(&lli->lli_och_mutex);
1015 spin_lock_init(&lli->lli_agl_lock);
1016 spin_lock_init(&lli->lli_layout_lock);
1017 ll_layout_version_set(lli, CL_LAYOUT_GEN_NONE);
1018 lli->lli_clob = NULL;
1020 init_rwsem(&lli->lli_xattrs_list_rwsem);
1021 mutex_init(&lli->lli_xattrs_enq_lock);
1023 LASSERT(lli->lli_vfs_inode.i_mode != 0);
1024 if (S_ISDIR(lli->lli_vfs_inode.i_mode)) {
1025 lli->lli_opendir_key = NULL;
1026 lli->lli_sai = NULL;
1027 spin_lock_init(&lli->lli_sa_lock);
1028 lli->lli_opendir_pid = 0;
1029 lli->lli_sa_enabled = 0;
1030 init_rwsem(&lli->lli_lsm_sem);
1032 mutex_init(&lli->lli_size_mutex);
1033 mutex_init(&lli->lli_setattr_mutex);
1034 lli->lli_symlink_name = NULL;
1035 ll_trunc_sem_init(&lli->lli_trunc_sem);
1036 range_lock_tree_init(&lli->lli_write_tree);
1037 init_rwsem(&lli->lli_glimpse_sem);
1038 lli->lli_glimpse_time = ktime_set(0, 0);
1039 INIT_LIST_HEAD(&lli->lli_agl_list);
1040 lli->lli_agl_index = 0;
1041 lli->lli_async_rc = 0;
1042 spin_lock_init(&lli->lli_heat_lock);
1043 obd_heat_clear(lli->lli_heat_instances, OBD_HEAT_COUNT);
1044 lli->lli_heat_flags = 0;
1045 mutex_init(&lli->lli_pcc_lock);
1046 lli->lli_pcc_state = PCC_STATE_FL_NONE;
1047 lli->lli_pcc_inode = NULL;
1048 lli->lli_pcc_dsflags = PCC_DATASET_INVALID;
1049 lli->lli_pcc_generation = 0;
1050 mutex_init(&lli->lli_group_mutex);
1051 lli->lli_group_users = 0;
1052 lli->lli_group_gid = 0;
1054 mutex_init(&lli->lli_layout_mutex);
1055 memset(lli->lli_jobid, 0, sizeof(lli->lli_jobid));
1058 #define MAX_STRING_SIZE 128
1060 #ifndef HAVE_SUPER_SETUP_BDI_NAME
1062 #define LSI_BDI_INITIALIZED 0x00400000
1064 #ifndef HAVE_BDI_CAP_MAP_COPY
1065 # define BDI_CAP_MAP_COPY 0
1068 static int super_setup_bdi_name(struct super_block *sb, char *fmt, ...)
1070 struct lustre_sb_info *lsi = s2lsi(sb);
1071 char buf[MAX_STRING_SIZE];
1075 err = bdi_init(&lsi->lsi_bdi);
1079 lsi->lsi_flags |= LSI_BDI_INITIALIZED;
1080 lsi->lsi_bdi.capabilities = BDI_CAP_MAP_COPY;
1081 lsi->lsi_bdi.name = "lustre";
1082 va_start(args, fmt);
1083 vsnprintf(buf, MAX_STRING_SIZE, fmt, args);
1085 err = bdi_register(&lsi->lsi_bdi, NULL, "%s", buf);
1088 sb->s_bdi = &lsi->lsi_bdi;
1092 #endif /* !HAVE_SUPER_SETUP_BDI_NAME */
1094 int ll_fill_super(struct super_block *sb)
1096 struct lustre_profile *lprof = NULL;
1097 struct lustre_sb_info *lsi = s2lsi(sb);
1098 struct ll_sb_info *sbi = NULL;
1099 char *dt = NULL, *md = NULL;
1100 char *profilenm = get_profile_name(sb);
1101 struct config_llog_instance *cfg;
1102 /* %p for void* in printf needs 16+2 characters: 0xffffffffffffffff */
1103 const int instlen = LUSTRE_MAXINSTANCE + 2;
1104 unsigned long cfg_instance = ll_get_cfg_instance(sb);
1105 char name[MAX_STRING_SIZE];
1114 /* for ASLR, to map between cfg_instance and hashed ptr */
1115 CDEBUG(D_VFSTRACE, "VFS Op: cfg_instance %s-%016lx (sb %p)\n",
1116 profilenm, cfg_instance, sb);
1120 GOTO(out_free_cfg, err = -ENOMEM);
1122 /* client additional sb info */
1123 lsi->lsi_llsbi = sbi = ll_init_sbi();
1125 GOTO(out_free_cfg, err = PTR_ERR(sbi));
1127 err = ll_options(lsi->lsi_lmd->lmd_opts, sbi);
1129 GOTO(out_free_cfg, err);
1131 /* kernel >= 2.6.38 store dentry operations in sb->s_d_op. */
1132 sb->s_d_op = &ll_d_ops;
1135 generate_random_uuid(uuid.b);
1136 snprintf(sbi->ll_sb_uuid.uuid, sizeof(sbi->ll_sb_uuid), "%pU", uuid.b);
1138 CDEBUG(D_CONFIG, "llite sb uuid: %s\n", sbi->ll_sb_uuid.uuid);
1141 len = strlen(profilenm);
1142 ptr = strrchr(profilenm, '-');
1143 if (ptr && (strcmp(ptr, "-client") == 0))
1146 if (len > LUSTRE_MAXFSNAME) {
1147 if (unlikely(len >= MAX_STRING_SIZE))
1148 len = MAX_STRING_SIZE - 1;
1149 strncpy(name, profilenm, len);
1151 err = -ENAMETOOLONG;
1152 CERROR("%s: fsname longer than %u characters: rc = %d\n",
1153 name, LUSTRE_MAXFSNAME, err);
1154 GOTO(out_free_cfg, err);
1156 strncpy(sbi->ll_fsname, profilenm, len);
1157 sbi->ll_fsname[len] = '\0';
1160 snprintf(name, sizeof(name), "%.*s-%016lx", len,
1161 profilenm, cfg_instance);
1163 err = super_setup_bdi_name(sb, "%s", name);
1165 GOTO(out_free_cfg, err);
1167 /* Call ll_debugfs_register_super() before lustre_process_log()
1168 * so that "llite.*.*" params can be processed correctly.
1170 err = ll_debugfs_register_super(sb, name);
1172 CERROR("%s: could not register mountpoint in llite: rc = %d\n",
1173 sbi->ll_fsname, err);
1177 /* The cfg_instance is a value unique to this super, in case some
1178 * joker tries to mount the same fs at two mount points.
1180 cfg->cfg_instance = cfg_instance;
1181 cfg->cfg_uuid = lsi->lsi_llsbi->ll_sb_uuid;
1182 cfg->cfg_callback = class_config_llog_handler;
1183 cfg->cfg_sub_clds = CONFIG_SUB_CLIENT;
1184 /* set up client obds */
1185 err = lustre_process_log(sb, profilenm, cfg);
1187 GOTO(out_debugfs, err);
1189 /* Profile set with LCFG_MOUNTOPT so we can find our mdc and osc obds */
1190 lprof = class_get_profile(profilenm);
1191 if (lprof == NULL) {
1192 LCONSOLE_ERROR_MSG(0x156, "The client profile '%s' could not be"
1193 " read from the MGS. Does that filesystem "
1194 "exist?\n", profilenm);
1195 GOTO(out_debugfs, err = -EINVAL);
1197 CDEBUG(D_CONFIG, "Found profile %s: mdc=%s osc=%s\n", profilenm,
1198 lprof->lp_md, lprof->lp_dt);
1200 dt_len = strlen(lprof->lp_dt) + instlen + 2;
1201 OBD_ALLOC(dt, dt_len);
1203 GOTO(out_profile, err = -ENOMEM);
1204 snprintf(dt, dt_len - 1, "%s-%016lx", lprof->lp_dt, cfg_instance);
1206 md_len = strlen(lprof->lp_md) + instlen + 2;
1207 OBD_ALLOC(md, md_len);
1209 GOTO(out_free_dt, err = -ENOMEM);
1210 snprintf(md, md_len - 1, "%s-%016lx", lprof->lp_md, cfg_instance);
1212 /* connections, registrations, sb setup */
1213 err = client_common_fill_super(sb, md, dt);
1215 GOTO(out_free_md, err);
1217 sbi->ll_client_common_fill_super_succeeded = 1;
1221 OBD_FREE(md, md_len);
1224 OBD_FREE(dt, dt_len);
1227 class_put_profile(lprof);
1230 ll_debugfs_unregister_super(sb);
1237 else if (sbi->ll_flags & LL_SBI_VERBOSE)
1238 LCONSOLE_WARN("Mounted %s\n", profilenm);
1240 } /* ll_fill_super */
1242 void ll_put_super(struct super_block *sb)
1244 struct config_llog_instance cfg, params_cfg;
1245 struct obd_device *obd;
1246 struct lustre_sb_info *lsi = s2lsi(sb);
1247 struct ll_sb_info *sbi = ll_s2sbi(sb);
1248 char *profilenm = get_profile_name(sb);
1249 unsigned long cfg_instance = ll_get_cfg_instance(sb);
1251 int next, force = 1, rc = 0;
1255 GOTO(out_no_sbi, 0);
1257 /* Should replace instance_id with something better for ASLR */
1258 CDEBUG(D_VFSTRACE, "VFS Op: cfg_instance %s-%016lx (sb %p)\n",
1259 profilenm, cfg_instance, sb);
1261 cfg.cfg_instance = cfg_instance;
1262 lustre_end_log(sb, profilenm, &cfg);
1264 params_cfg.cfg_instance = cfg_instance;
1265 lustre_end_log(sb, PARAMS_FILENAME, ¶ms_cfg);
1267 if (sbi->ll_md_exp) {
1268 obd = class_exp2obd(sbi->ll_md_exp);
1270 force = obd->obd_force;
1273 /* Wait for unstable pages to be committed to stable storage */
1275 rc = l_wait_event_abortable(
1276 sbi->ll_cache->ccc_unstable_waitq,
1277 atomic_long_read(&sbi->ll_cache->ccc_unstable_nr) == 0);
1280 ccc_count = atomic_long_read(&sbi->ll_cache->ccc_unstable_nr);
1281 if (force == 0 && rc != -ERESTARTSYS)
1282 LASSERTF(ccc_count == 0, "count: %li\n", ccc_count);
1284 /* We need to set force before the lov_disconnect in
1285 * lustre_common_put_super, since l_d cleans up osc's as well.
1289 while ((obd = class_devices_in_group(&sbi->ll_sb_uuid,
1291 obd->obd_force = force;
1295 if (sbi->ll_client_common_fill_super_succeeded) {
1296 /* Only if client_common_fill_super succeeded */
1297 client_common_put_super(sb);
1301 while ((obd = class_devices_in_group(&sbi->ll_sb_uuid, &next)))
1302 class_manual_cleanup(obd);
1304 if (sbi->ll_flags & LL_SBI_VERBOSE)
1305 LCONSOLE_WARN("Unmounted %s\n", profilenm ? profilenm : "");
1308 class_del_profile(profilenm);
1310 #ifndef HAVE_SUPER_SETUP_BDI_NAME
1311 if (lsi->lsi_flags & LSI_BDI_INITIALIZED) {
1312 bdi_destroy(&lsi->lsi_bdi);
1313 lsi->lsi_flags &= ~LSI_BDI_INITIALIZED;
1318 lsi->lsi_llsbi = NULL;
1320 lustre_common_put_super(sb);
1322 cl_env_cache_purge(~0);
1324 module_put(THIS_MODULE);
1327 } /* client_put_super */
1329 struct inode *ll_inode_from_resource_lock(struct ldlm_lock *lock)
1331 struct inode *inode = NULL;
1333 /* NOTE: we depend on atomic igrab() -bzzz */
1334 lock_res_and_lock(lock);
1335 if (lock->l_resource->lr_lvb_inode) {
1336 struct ll_inode_info * lli;
1337 lli = ll_i2info(lock->l_resource->lr_lvb_inode);
1338 if (lli->lli_inode_magic == LLI_INODE_MAGIC) {
1339 inode = igrab(lock->l_resource->lr_lvb_inode);
1341 inode = lock->l_resource->lr_lvb_inode;
1342 LDLM_DEBUG_LIMIT(inode->i_state & I_FREEING ? D_INFO :
1343 D_WARNING, lock, "lr_lvb_inode %p is "
1344 "bogus: magic %08x",
1345 lock->l_resource->lr_lvb_inode,
1346 lli->lli_inode_magic);
1350 unlock_res_and_lock(lock);
1354 void ll_dir_clear_lsm_md(struct inode *inode)
1356 struct ll_inode_info *lli = ll_i2info(inode);
1358 LASSERT(S_ISDIR(inode->i_mode));
1360 if (lli->lli_lsm_md) {
1361 lmv_free_memmd(lli->lli_lsm_md);
1362 lli->lli_lsm_md = NULL;
1365 if (lli->lli_default_lsm_md) {
1366 lmv_free_memmd(lli->lli_default_lsm_md);
1367 lli->lli_default_lsm_md = NULL;
1371 static struct inode *ll_iget_anon_dir(struct super_block *sb,
1372 const struct lu_fid *fid,
1373 struct lustre_md *md)
1375 struct ll_sb_info *sbi = ll_s2sbi(sb);
1376 struct mdt_body *body = md->body;
1377 struct inode *inode;
1381 ino = cl_fid_build_ino(fid, sbi->ll_flags & LL_SBI_32BIT_API);
1382 inode = iget_locked(sb, ino);
1383 if (inode == NULL) {
1384 CERROR("%s: failed get simple inode "DFID": rc = -ENOENT\n",
1385 sbi->ll_fsname, PFID(fid));
1386 RETURN(ERR_PTR(-ENOENT));
1389 if (inode->i_state & I_NEW) {
1390 struct ll_inode_info *lli = ll_i2info(inode);
1391 struct lmv_stripe_md *lsm = md->lmv;
1393 inode->i_mode = (inode->i_mode & ~S_IFMT) |
1394 (body->mbo_mode & S_IFMT);
1395 LASSERTF(S_ISDIR(inode->i_mode), "Not slave inode "DFID"\n",
1398 inode->i_mtime.tv_sec = 0;
1399 inode->i_atime.tv_sec = 0;
1400 inode->i_ctime.tv_sec = 0;
1403 #ifdef HAVE_BACKING_DEV_INFO
1404 /* initializing backing dev info. */
1405 inode->i_mapping->backing_dev_info =
1406 &s2lsi(inode->i_sb)->lsi_bdi;
1408 inode->i_op = &ll_dir_inode_operations;
1409 inode->i_fop = &ll_dir_operations;
1410 lli->lli_fid = *fid;
1413 LASSERT(lsm != NULL);
1414 /* master object FID */
1415 lli->lli_pfid = body->mbo_fid1;
1416 CDEBUG(D_INODE, "lli %p slave "DFID" master "DFID"\n",
1417 lli, PFID(fid), PFID(&lli->lli_pfid));
1418 unlock_new_inode(inode);
1424 static int ll_init_lsm_md(struct inode *inode, struct lustre_md *md)
1427 struct lmv_stripe_md *lsm = md->lmv;
1428 struct ll_inode_info *lli = ll_i2info(inode);
1431 LASSERT(lsm != NULL);
1433 CDEBUG(D_INODE, "%s: "DFID" set dir layout:\n",
1434 ll_i2sbi(inode)->ll_fsname, PFID(&lli->lli_fid));
1435 lsm_md_dump(D_INODE, lsm);
1437 if (!lmv_dir_striped(lsm))
1440 /* XXX sigh, this lsm_root initialization should be in
1441 * LMV layer, but it needs ll_iget right now, so we
1442 * put this here right now. */
1443 for (i = 0; i < lsm->lsm_md_stripe_count; i++) {
1444 fid = &lsm->lsm_md_oinfo[i].lmo_fid;
1445 LASSERT(lsm->lsm_md_oinfo[i].lmo_root == NULL);
1447 if (!fid_is_sane(fid))
1450 /* Unfortunately ll_iget will call ll_update_inode,
1451 * where the initialization of slave inode is slightly
1452 * different, so it reset lsm_md to NULL to avoid
1453 * initializing lsm for slave inode. */
1454 lsm->lsm_md_oinfo[i].lmo_root =
1455 ll_iget_anon_dir(inode->i_sb, fid, md);
1456 if (IS_ERR(lsm->lsm_md_oinfo[i].lmo_root)) {
1457 int rc = PTR_ERR(lsm->lsm_md_oinfo[i].lmo_root);
1459 lsm->lsm_md_oinfo[i].lmo_root = NULL;
1461 iput(lsm->lsm_md_oinfo[i].lmo_root);
1462 lsm->lsm_md_oinfo[i].lmo_root = NULL;
1468 lli->lli_lsm_md = lsm;
1473 static void ll_update_default_lsm_md(struct inode *inode, struct lustre_md *md)
1475 struct ll_inode_info *lli = ll_i2info(inode);
1477 if (!md->default_lmv) {
1478 /* clear default lsm */
1479 if (lli->lli_default_lsm_md) {
1480 down_write(&lli->lli_lsm_sem);
1481 if (lli->lli_default_lsm_md) {
1482 lmv_free_memmd(lli->lli_default_lsm_md);
1483 lli->lli_default_lsm_md = NULL;
1485 up_write(&lli->lli_lsm_sem);
1487 } else if (lli->lli_default_lsm_md) {
1488 /* update default lsm if it changes */
1489 down_read(&lli->lli_lsm_sem);
1490 if (lli->lli_default_lsm_md &&
1491 !lsm_md_eq(lli->lli_default_lsm_md, md->default_lmv)) {
1492 up_read(&lli->lli_lsm_sem);
1493 down_write(&lli->lli_lsm_sem);
1494 if (lli->lli_default_lsm_md)
1495 lmv_free_memmd(lli->lli_default_lsm_md);
1496 lli->lli_default_lsm_md = md->default_lmv;
1497 lsm_md_dump(D_INODE, md->default_lmv);
1498 md->default_lmv = NULL;
1499 up_write(&lli->lli_lsm_sem);
1501 up_read(&lli->lli_lsm_sem);
1504 /* init default lsm */
1505 down_write(&lli->lli_lsm_sem);
1506 lli->lli_default_lsm_md = md->default_lmv;
1507 lsm_md_dump(D_INODE, md->default_lmv);
1508 md->default_lmv = NULL;
1509 up_write(&lli->lli_lsm_sem);
1513 static int ll_update_lsm_md(struct inode *inode, struct lustre_md *md)
1515 struct ll_inode_info *lli = ll_i2info(inode);
1516 struct lmv_stripe_md *lsm = md->lmv;
1517 struct cl_attr *attr;
1522 LASSERT(S_ISDIR(inode->i_mode));
1523 CDEBUG(D_INODE, "update lsm %p of "DFID"\n", lli->lli_lsm_md,
1524 PFID(ll_inode2fid(inode)));
1526 /* update default LMV */
1527 if (md->default_lmv)
1528 ll_update_default_lsm_md(inode, md);
1531 * no striped information from request, lustre_md from req does not
1532 * include stripeEA, see ll_md_setattr()
1538 * normally dir layout doesn't change, only take read lock to check
1539 * that to avoid blocking other MD operations.
1541 down_read(&lli->lli_lsm_sem);
1543 /* some current lookup initialized lsm, and unchanged */
1544 if (lli->lli_lsm_md && lsm_md_eq(lli->lli_lsm_md, lsm))
1545 GOTO(unlock, rc = 0);
1547 /* if dir layout doesn't match, check whether version is increased,
1548 * which means layout is changed, this happens in dir split/merge and
1551 * foreign LMV should not change.
1553 if (lli->lli_lsm_md && lmv_dir_striped(lli->lli_lsm_md) &&
1554 lsm->lsm_md_layout_version <=
1555 lli->lli_lsm_md->lsm_md_layout_version) {
1556 CERROR("%s: "DFID" dir layout mismatch:\n",
1557 ll_i2sbi(inode)->ll_fsname, PFID(&lli->lli_fid));
1558 lsm_md_dump(D_ERROR, lli->lli_lsm_md);
1559 lsm_md_dump(D_ERROR, lsm);
1560 GOTO(unlock, rc = -EINVAL);
1563 up_read(&lli->lli_lsm_sem);
1564 down_write(&lli->lli_lsm_sem);
1565 /* clear existing lsm */
1566 if (lli->lli_lsm_md) {
1567 lmv_free_memmd(lli->lli_lsm_md);
1568 lli->lli_lsm_md = NULL;
1571 rc = ll_init_lsm_md(inode, md);
1572 up_write(&lli->lli_lsm_sem);
1577 /* set md->lmv to NULL, so the following free lustre_md will not free
1582 /* md_merge_attr() may take long, since lsm is already set, switch to
1585 down_read(&lli->lli_lsm_sem);
1587 if (!lmv_dir_striped(lli->lli_lsm_md))
1588 GOTO(unlock, rc = 0);
1590 OBD_ALLOC_PTR(attr);
1592 GOTO(unlock, rc = -ENOMEM);
1594 /* validate the lsm */
1595 rc = md_merge_attr(ll_i2mdexp(inode), &lli->lli_fid, lli->lli_lsm_md,
1596 attr, ll_md_blocking_ast);
1598 if (md->body->mbo_valid & OBD_MD_FLNLINK)
1599 md->body->mbo_nlink = attr->cat_nlink;
1600 if (md->body->mbo_valid & OBD_MD_FLSIZE)
1601 md->body->mbo_size = attr->cat_size;
1602 if (md->body->mbo_valid & OBD_MD_FLATIME)
1603 md->body->mbo_atime = attr->cat_atime;
1604 if (md->body->mbo_valid & OBD_MD_FLCTIME)
1605 md->body->mbo_ctime = attr->cat_ctime;
1606 if (md->body->mbo_valid & OBD_MD_FLMTIME)
1607 md->body->mbo_mtime = attr->cat_mtime;
1613 up_read(&lli->lli_lsm_sem);
1618 void ll_clear_inode(struct inode *inode)
1620 struct ll_inode_info *lli = ll_i2info(inode);
1621 struct ll_sb_info *sbi = ll_i2sbi(inode);
1625 CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p)\n",
1626 PFID(ll_inode2fid(inode)), inode);
1628 if (S_ISDIR(inode->i_mode)) {
1629 /* these should have been cleared in ll_file_release */
1630 LASSERT(lli->lli_opendir_key == NULL);
1631 LASSERT(lli->lli_sai == NULL);
1632 LASSERT(lli->lli_opendir_pid == 0);
1634 pcc_inode_free(inode);
1637 md_null_inode(sbi->ll_md_exp, ll_inode2fid(inode));
1639 LASSERT(!lli->lli_open_fd_write_count);
1640 LASSERT(!lli->lli_open_fd_read_count);
1641 LASSERT(!lli->lli_open_fd_exec_count);
1643 if (lli->lli_mds_write_och)
1644 ll_md_real_close(inode, FMODE_WRITE);
1645 if (lli->lli_mds_exec_och)
1646 ll_md_real_close(inode, FMODE_EXEC);
1647 if (lli->lli_mds_read_och)
1648 ll_md_real_close(inode, FMODE_READ);
1650 if (S_ISLNK(inode->i_mode) && lli->lli_symlink_name) {
1651 OBD_FREE(lli->lli_symlink_name,
1652 strlen(lli->lli_symlink_name) + 1);
1653 lli->lli_symlink_name = NULL;
1656 ll_xattr_cache_destroy(inode);
1658 forget_all_cached_acls(inode);
1660 lli->lli_inode_magic = LLI_INODE_DEAD;
1662 if (S_ISDIR(inode->i_mode))
1663 ll_dir_clear_lsm_md(inode);
1664 else if (S_ISREG(inode->i_mode) && !is_bad_inode(inode))
1665 LASSERT(list_empty(&lli->lli_agl_list));
1668 * XXX This has to be done before lsm is freed below, because
1669 * cl_object still uses inode lsm.
1671 cl_inode_fini(inode);
1673 llcrypt_put_encryption_info(inode);
1678 static int ll_md_setattr(struct dentry *dentry, struct md_op_data *op_data)
1680 struct lustre_md md;
1681 struct inode *inode = dentry->d_inode;
1682 struct ll_sb_info *sbi = ll_i2sbi(inode);
1683 struct ptlrpc_request *request = NULL;
1687 op_data = ll_prep_md_op_data(op_data, inode, NULL, NULL, 0, 0,
1688 LUSTRE_OPC_ANY, NULL);
1689 if (IS_ERR(op_data))
1690 RETURN(PTR_ERR(op_data));
1692 rc = md_setattr(sbi->ll_md_exp, op_data, NULL, 0, &request);
1694 ptlrpc_req_finished(request);
1695 if (rc == -ENOENT) {
1697 /* Unlinked special device node? Or just a race?
1698 * Pretend we done everything. */
1699 if (!S_ISREG(inode->i_mode) &&
1700 !S_ISDIR(inode->i_mode)) {
1701 ia_valid = op_data->op_attr.ia_valid;
1702 op_data->op_attr.ia_valid &= ~TIMES_SET_FLAGS;
1703 rc = simple_setattr(dentry, &op_data->op_attr);
1704 op_data->op_attr.ia_valid = ia_valid;
1706 } else if (rc != -EPERM && rc != -EACCES && rc != -ETXTBSY) {
1707 CERROR("md_setattr fails: rc = %d\n", rc);
1712 rc = md_get_lustre_md(sbi->ll_md_exp, request, sbi->ll_dt_exp,
1713 sbi->ll_md_exp, &md);
1715 ptlrpc_req_finished(request);
1719 ia_valid = op_data->op_attr.ia_valid;
1720 /* inode size will be in ll_setattr_ost, can't do it now since dirty
1721 * cache is not cleared yet. */
1722 op_data->op_attr.ia_valid &= ~(TIMES_SET_FLAGS | ATTR_SIZE);
1723 if (S_ISREG(inode->i_mode))
1725 rc = simple_setattr(dentry, &op_data->op_attr);
1726 if (S_ISREG(inode->i_mode))
1727 inode_unlock(inode);
1728 op_data->op_attr.ia_valid = ia_valid;
1730 rc = ll_update_inode(inode, &md);
1731 ptlrpc_req_finished(request);
1737 * Zero portion of page that is part of @inode.
1738 * This implies, if necessary:
1739 * - taking cl_lock on range corresponding to concerned page
1740 * - grabbing vm page
1741 * - associating cl_page
1742 * - proceeding to clio read
1743 * - zeroing range in page
1744 * - proceeding to cl_page flush
1745 * - releasing cl_lock
1747 * \param[in] inode inode
1748 * \param[in] index page index
1749 * \param[in] offset offset in page to start zero from
1750 * \param[in] len len to zero
1752 * \retval 0 on success
1753 * \retval negative errno on failure
1755 int ll_io_zero_page(struct inode *inode, pgoff_t index, pgoff_t offset,
1758 struct ll_inode_info *lli = ll_i2info(inode);
1759 struct cl_object *clob = lli->lli_clob;
1761 struct lu_env *env = NULL;
1762 struct cl_io *io = NULL;
1763 struct cl_page *clpage = NULL;
1764 struct page *vmpage = NULL;
1765 unsigned from = index << PAGE_SHIFT;
1766 struct cl_lock *lock = NULL;
1767 struct cl_lock_descr *descr = NULL;
1768 struct cl_2queue *queue = NULL;
1769 struct cl_sync_io *anchor = NULL;
1770 bool holdinglock = false;
1771 bool lockedbymyself = true;
1776 env = cl_env_get(&refcheck);
1778 RETURN(PTR_ERR(env));
1780 io = vvp_env_thread_io(env);
1782 rc = cl_io_rw_init(env, io, CIT_WRITE, from, PAGE_SIZE);
1786 lock = vvp_env_lock(env);
1787 descr = &lock->cll_descr;
1788 descr->cld_obj = io->ci_obj;
1789 descr->cld_start = cl_index(io->ci_obj, from);
1790 descr->cld_end = cl_index(io->ci_obj, from + PAGE_SIZE - 1);
1791 descr->cld_mode = CLM_WRITE;
1792 descr->cld_enq_flags = CEF_MUST | CEF_NONBLOCK;
1794 /* request lock for page */
1795 rc = cl_lock_request(env, io, lock);
1796 /* -ECANCELED indicates a matching lock with a different extent
1797 * was already present, and -EEXIST indicates a matching lock
1798 * on exactly the same extent was already present.
1799 * In both cases it means we are covered.
1801 if (rc == -ECANCELED || rc == -EEXIST)
1809 vmpage = grab_cache_page_nowait(inode->i_mapping, index);
1811 GOTO(rellock, rc = -EOPNOTSUPP);
1813 if (!PageDirty(vmpage)) {
1814 /* associate cl_page */
1815 clpage = cl_page_find(env, clob, vmpage->index,
1816 vmpage, CPT_CACHEABLE);
1818 GOTO(pagefini, rc = PTR_ERR(clpage));
1820 cl_page_assume(env, io, clpage);
1823 if (!PageUptodate(vmpage) && !PageDirty(vmpage) &&
1824 !PageWriteback(vmpage)) {
1826 /* set PagePrivate2 to detect special case of empty page
1827 * in osc_brw_fini_request()
1829 SetPagePrivate2(vmpage);
1830 rc = ll_io_read_page(env, io, clpage, NULL);
1831 if (!PagePrivate2(vmpage))
1832 /* PagePrivate2 was cleared in osc_brw_fini_request()
1833 * meaning we read an empty page. In this case, in order
1834 * to avoid allocating unnecessary block in truncated
1835 * file, we must not zero and write as below. Subsequent
1836 * server-side truncate will handle things correctly.
1838 GOTO(clpfini, rc = 0);
1839 ClearPagePrivate2(vmpage);
1842 lockedbymyself = trylock_page(vmpage);
1843 cl_page_assume(env, io, clpage);
1846 /* zero range in page */
1847 zero_user(vmpage, offset, len);
1849 if (holdinglock && clpage) {
1850 /* explicitly write newly modified page */
1851 queue = &io->ci_queue;
1852 cl_2queue_init(queue);
1853 anchor = &vvp_env_info(env)->vti_anchor;
1854 cl_sync_io_init(anchor, 1);
1855 clpage->cp_sync_io = anchor;
1856 cl_2queue_add(queue, clpage);
1857 rc = cl_io_submit_rw(env, io, CRT_WRITE, queue);
1859 GOTO(queuefini1, rc);
1860 rc = cl_sync_io_wait(env, anchor, 0);
1862 GOTO(queuefini2, rc);
1863 cl_page_assume(env, io, clpage);
1866 cl_2queue_discard(env, io, queue);
1868 cl_2queue_disown(env, io, queue);
1869 cl_2queue_fini(env, queue);
1874 cl_page_put(env, clpage);
1876 if (lockedbymyself) {
1877 unlock_page(vmpage);
1882 cl_lock_release(env, lock);
1884 cl_io_fini(env, io);
1887 cl_env_put(env, &refcheck);
1892 /* If this inode has objects allocated to it (lsm != NULL), then the OST
1893 * object(s) determine the file size and mtime. Otherwise, the MDS will
1894 * keep these values until such a time that objects are allocated for it.
1895 * We do the MDS operations first, as it is checking permissions for us.
1896 * We don't to the MDS RPC if there is nothing that we want to store there,
1897 * otherwise there is no harm in updating mtime/atime on the MDS if we are
1898 * going to do an RPC anyways.
1900 * If we are doing a truncate, we will send the mtime and ctime updates
1901 * to the OST with the punch RPC, otherwise we do an explicit setattr RPC.
1902 * I don't believe it is possible to get e.g. ATTR_MTIME_SET and ATTR_SIZE
1905 * In case of HSMimport, we only set attr on MDS.
1907 int ll_setattr_raw(struct dentry *dentry, struct iattr *attr,
1908 enum op_xvalid xvalid, bool hsm_import)
1910 struct inode *inode = dentry->d_inode;
1911 struct ll_inode_info *lli = ll_i2info(inode);
1912 struct md_op_data *op_data = NULL;
1913 ktime_t kstart = ktime_get();
1918 CDEBUG(D_VFSTRACE, "%s: setattr inode "DFID"(%p) from %llu to %llu, "
1919 "valid %x, hsm_import %d\n",
1920 ll_i2sbi(inode)->ll_fsname, PFID(&lli->lli_fid),
1921 inode, i_size_read(inode), attr->ia_size, attr->ia_valid,
1924 if (attr->ia_valid & ATTR_SIZE) {
1925 /* Check new size against VFS/VM file size limit and rlimit */
1926 rc = inode_newsize_ok(inode, attr->ia_size);
1930 /* The maximum Lustre file size is variable, based on the
1931 * OST maximum object size and number of stripes. This
1932 * needs another check in addition to the VFS check above. */
1933 if (attr->ia_size > ll_file_maxbytes(inode)) {
1934 CDEBUG(D_INODE,"file "DFID" too large %llu > %llu\n",
1935 PFID(&lli->lli_fid), attr->ia_size,
1936 ll_file_maxbytes(inode));
1940 attr->ia_valid |= ATTR_MTIME | ATTR_CTIME;
1943 /* POSIX: check before ATTR_*TIME_SET set (from inode_change_ok) */
1944 if (attr->ia_valid & TIMES_SET_FLAGS) {
1945 if ((!uid_eq(current_fsuid(), inode->i_uid)) &&
1946 !cfs_capable(CFS_CAP_FOWNER))
1950 /* We mark all of the fields "set" so MDS/OST does not re-set them */
1951 if (!(xvalid & OP_XVALID_CTIME_SET) &&
1952 (attr->ia_valid & ATTR_CTIME)) {
1953 attr->ia_ctime = current_time(inode);
1954 xvalid |= OP_XVALID_CTIME_SET;
1956 if (!(attr->ia_valid & ATTR_ATIME_SET) &&
1957 (attr->ia_valid & ATTR_ATIME)) {
1958 attr->ia_atime = current_time(inode);
1959 attr->ia_valid |= ATTR_ATIME_SET;
1961 if (!(attr->ia_valid & ATTR_MTIME_SET) &&
1962 (attr->ia_valid & ATTR_MTIME)) {
1963 attr->ia_mtime = current_time(inode);
1964 attr->ia_valid |= ATTR_MTIME_SET;
1967 if (attr->ia_valid & (ATTR_MTIME | ATTR_CTIME))
1968 CDEBUG(D_INODE, "setting mtime %lld, ctime %lld, now = %lld\n",
1969 (s64)attr->ia_mtime.tv_sec, (s64)attr->ia_ctime.tv_sec,
1970 ktime_get_real_seconds());
1972 if (S_ISREG(inode->i_mode))
1973 inode_unlock(inode);
1975 /* We always do an MDS RPC, even if we're only changing the size;
1976 * only the MDS knows whether truncate() should fail with -ETXTBUSY */
1978 OBD_ALLOC_PTR(op_data);
1979 if (op_data == NULL)
1980 GOTO(out, rc = -ENOMEM);
1982 if (!hsm_import && attr->ia_valid & ATTR_SIZE) {
1983 /* If we are changing file size, file content is
1984 * modified, flag it.
1986 xvalid |= OP_XVALID_OWNEROVERRIDE;
1987 op_data->op_bias |= MDS_DATA_MODIFIED;
1988 ll_file_clear_flag(lli, LLIF_DATA_MODIFIED);
1991 if (attr->ia_valid & ATTR_FILE) {
1992 struct ll_file_data *fd = attr->ia_file->private_data;
1994 if (fd->fd_lease_och)
1995 op_data->op_bias |= MDS_TRUNC_KEEP_LEASE;
1998 op_data->op_attr = *attr;
1999 op_data->op_xvalid = xvalid;
2001 rc = ll_md_setattr(dentry, op_data);
2005 if (!S_ISREG(inode->i_mode) || hsm_import)
2008 if (attr->ia_valid & (ATTR_SIZE | ATTR_ATIME | ATTR_ATIME_SET |
2009 ATTR_MTIME | ATTR_MTIME_SET | ATTR_CTIME) ||
2010 xvalid & OP_XVALID_CTIME_SET) {
2011 bool cached = false;
2013 rc = pcc_inode_setattr(inode, attr, &cached);
2016 CERROR("%s: PCC inode "DFID" setattr failed: "
2018 ll_i2sbi(inode)->ll_fsname,
2019 PFID(&lli->lli_fid), rc);
2023 unsigned int flags = 0;
2025 /* For truncate and utimes sending attributes to OSTs,
2026 * setting mtime/atime to the past will be performed
2027 * under PW [0:EOF] extent lock (new_size:EOF for
2028 * truncate). It may seem excessive to send mtime/atime
2029 * updates to OSTs when not setting times to past, but
2030 * it is necessary due to possible time
2031 * de-synchronization between MDT inode and OST objects
2033 if (S_ISREG(inode->i_mode) && IS_ENCRYPTED(inode) &&
2034 attr->ia_valid & ATTR_SIZE) {
2035 xvalid |= OP_XVALID_FLAGS;
2036 flags = LUSTRE_ENCRYPT_FL;
2037 if (attr->ia_size & ~PAGE_MASK) {
2039 attr->ia_size & (PAGE_SIZE - 1);
2041 rc = ll_io_zero_page(inode,
2042 attr->ia_size >> PAGE_SHIFT,
2043 offset, PAGE_SIZE - offset);
2048 rc = cl_setattr_ost(lli->lli_clob, attr, xvalid, flags);
2052 /* If the file was restored, it needs to set dirty flag.
2054 * We've already sent MDS_DATA_MODIFIED flag in
2055 * ll_md_setattr() for truncate. However, the MDT refuses to
2056 * set the HS_DIRTY flag on released files, so we have to set
2057 * it again if the file has been restored. Please check how
2058 * LLIF_DATA_MODIFIED is set in vvp_io_setattr_fini().
2060 * Please notice that if the file is not released, the previous
2061 * MDS_DATA_MODIFIED has taken effect and usually
2062 * LLIF_DATA_MODIFIED is not set(see vvp_io_setattr_fini()).
2063 * This way we can save an RPC for common open + trunc
2065 if (ll_file_test_and_clear_flag(lli, LLIF_DATA_MODIFIED)) {
2066 struct hsm_state_set hss = {
2067 .hss_valid = HSS_SETMASK,
2068 .hss_setmask = HS_DIRTY,
2072 rc2 = ll_hsm_state_set(inode, &hss);
2073 /* truncate and write can happen at the same time, so that
2074 * the file can be set modified even though the file is not
2075 * restored from released state, and ll_hsm_state_set() is
2076 * not applicable for the file, and rc2 < 0 is normal in this
2079 CDEBUG(D_INFO, DFID "HSM set dirty failed: rc2 = %d\n",
2080 PFID(ll_inode2fid(inode)), rc2);
2085 if (op_data != NULL)
2086 ll_finish_md_op_data(op_data);
2088 if (S_ISREG(inode->i_mode)) {
2090 if ((attr->ia_valid & ATTR_SIZE) && !hsm_import)
2091 inode_dio_wait(inode);
2092 /* Once we've got the i_mutex, it's safe to set the S_NOSEC
2093 * flag. ll_update_inode (called from ll_md_setattr), clears
2094 * inode flags, so there is a gap where S_NOSEC is not set.
2095 * This can cause a writer to take the i_mutex unnecessarily,
2096 * but this is safe to do and should be rare. */
2097 inode_has_no_xattr(inode);
2101 ll_stats_ops_tally(ll_i2sbi(inode), attr->ia_valid & ATTR_SIZE ?
2102 LPROC_LL_TRUNC : LPROC_LL_SETATTR,
2103 ktime_us_delta(ktime_get(), kstart));
2108 int ll_setattr(struct dentry *de, struct iattr *attr)
2110 int mode = de->d_inode->i_mode;
2111 enum op_xvalid xvalid = 0;
2114 rc = llcrypt_prepare_setattr(de, attr);
2118 if ((attr->ia_valid & (ATTR_CTIME|ATTR_SIZE|ATTR_MODE)) ==
2119 (ATTR_CTIME|ATTR_SIZE|ATTR_MODE))
2120 xvalid |= OP_XVALID_OWNEROVERRIDE;
2122 if (((attr->ia_valid & (ATTR_MODE|ATTR_FORCE|ATTR_SIZE)) ==
2123 (ATTR_SIZE|ATTR_MODE)) &&
2124 (((mode & S_ISUID) && !(attr->ia_mode & S_ISUID)) ||
2125 (((mode & (S_ISGID|S_IXGRP)) == (S_ISGID|S_IXGRP)) &&
2126 !(attr->ia_mode & S_ISGID))))
2127 attr->ia_valid |= ATTR_FORCE;
2129 if ((attr->ia_valid & ATTR_MODE) &&
2131 !(attr->ia_mode & S_ISUID) &&
2132 !(attr->ia_valid & ATTR_KILL_SUID))
2133 attr->ia_valid |= ATTR_KILL_SUID;
2135 if ((attr->ia_valid & ATTR_MODE) &&
2136 ((mode & (S_ISGID|S_IXGRP)) == (S_ISGID|S_IXGRP)) &&
2137 !(attr->ia_mode & S_ISGID) &&
2138 !(attr->ia_valid & ATTR_KILL_SGID))
2139 attr->ia_valid |= ATTR_KILL_SGID;
2141 return ll_setattr_raw(de, attr, xvalid, false);
2144 int ll_statfs_internal(struct ll_sb_info *sbi, struct obd_statfs *osfs,
2147 struct obd_statfs obd_osfs = { 0 };
2152 max_age = ktime_get_seconds() - sbi->ll_statfs_max_age;
2154 if (sbi->ll_flags & LL_SBI_LAZYSTATFS)
2155 flags |= OBD_STATFS_NODELAY;
2157 rc = obd_statfs(NULL, sbi->ll_md_exp, osfs, max_age, flags);
2161 osfs->os_type = LL_SUPER_MAGIC;
2163 CDEBUG(D_SUPER, "MDC blocks %llu/%llu objects %llu/%llu\n",
2164 osfs->os_bavail, osfs->os_blocks, osfs->os_ffree, osfs->os_files);
2166 if (osfs->os_state & OS_STATFS_SUM)
2169 rc = obd_statfs(NULL, sbi->ll_dt_exp, &obd_osfs, max_age, flags);
2170 if (rc) /* Possibly a filesystem with no OSTs. Report MDT totals. */
2173 CDEBUG(D_SUPER, "OSC blocks %llu/%llu objects %llu/%llu\n",
2174 obd_osfs.os_bavail, obd_osfs.os_blocks, obd_osfs.os_ffree,
2177 osfs->os_bsize = obd_osfs.os_bsize;
2178 osfs->os_blocks = obd_osfs.os_blocks;
2179 osfs->os_bfree = obd_osfs.os_bfree;
2180 osfs->os_bavail = obd_osfs.os_bavail;
2182 /* If we have _some_ OSTs, but don't have as many free objects on the
2183 * OSTs as inodes on the MDTs, reduce the reported number of inodes
2184 * to compensate, so that the "inodes in use" number is correct.
2185 * This should be kept in sync with lod_statfs() behaviour.
2187 if (obd_osfs.os_files && obd_osfs.os_ffree < osfs->os_ffree) {
2188 osfs->os_files = (osfs->os_files - osfs->os_ffree) +
2190 osfs->os_ffree = obd_osfs.os_ffree;
2197 int ll_statfs(struct dentry *de, struct kstatfs *sfs)
2199 struct super_block *sb = de->d_sb;
2200 struct obd_statfs osfs;
2201 __u64 fsid = huge_encode_dev(sb->s_dev);
2202 ktime_t kstart = ktime_get();
2205 CDEBUG(D_VFSTRACE, "VFS Op:sb=%s (%p)\n", sb->s_id, sb);
2207 /* Some amount of caching on the client is allowed */
2208 rc = ll_statfs_internal(ll_s2sbi(sb), &osfs, OBD_STATFS_SUM);
2212 statfs_unpack(sfs, &osfs);
2214 /* We need to downshift for all 32-bit kernels, because we can't
2215 * tell if the kernel is being called via sys_statfs64() or not.
2216 * Stop before overflowing f_bsize - in which case it is better
2217 * to just risk EOVERFLOW if caller is using old sys_statfs(). */
2218 if (sizeof(long) < 8) {
2219 while (osfs.os_blocks > ~0UL && sfs->f_bsize < 0x40000000) {
2222 osfs.os_blocks >>= 1;
2223 osfs.os_bfree >>= 1;
2224 osfs.os_bavail >>= 1;
2228 sfs->f_blocks = osfs.os_blocks;
2229 sfs->f_bfree = osfs.os_bfree;
2230 sfs->f_bavail = osfs.os_bavail;
2231 sfs->f_fsid.val[0] = (__u32)fsid;
2232 sfs->f_fsid.val[1] = (__u32)(fsid >> 32);
2234 ll_stats_ops_tally(ll_s2sbi(sb), LPROC_LL_STATFS,
2235 ktime_us_delta(ktime_get(), kstart));
2240 void ll_inode_size_lock(struct inode *inode)
2242 struct ll_inode_info *lli;
2244 LASSERT(!S_ISDIR(inode->i_mode));
2246 lli = ll_i2info(inode);
2247 mutex_lock(&lli->lli_size_mutex);
2250 void ll_inode_size_unlock(struct inode *inode)
2252 struct ll_inode_info *lli;
2254 lli = ll_i2info(inode);
2255 mutex_unlock(&lli->lli_size_mutex);
2258 void ll_update_inode_flags(struct inode *inode, int ext_flags)
2260 /* do not clear encryption flag */
2261 ext_flags |= ll_inode_to_ext_flags(inode->i_flags) & LUSTRE_ENCRYPT_FL;
2262 inode->i_flags = ll_ext_to_inode_flags(ext_flags);
2263 if (ext_flags & LUSTRE_PROJINHERIT_FL)
2264 ll_file_set_flag(ll_i2info(inode), LLIF_PROJECT_INHERIT);
2266 ll_file_clear_flag(ll_i2info(inode), LLIF_PROJECT_INHERIT);
2269 int ll_update_inode(struct inode *inode, struct lustre_md *md)
2271 struct ll_inode_info *lli = ll_i2info(inode);
2272 struct mdt_body *body = md->body;
2273 struct ll_sb_info *sbi = ll_i2sbi(inode);
2276 if (body->mbo_valid & OBD_MD_FLEASIZE) {
2277 rc = cl_file_inode_init(inode, md);
2282 if (S_ISDIR(inode->i_mode)) {
2283 rc = ll_update_lsm_md(inode, md);
2288 if (body->mbo_valid & OBD_MD_FLACL)
2289 lli_replace_acl(lli, md);
2291 inode->i_ino = cl_fid_build_ino(&body->mbo_fid1,
2292 sbi->ll_flags & LL_SBI_32BIT_API);
2293 inode->i_generation = cl_fid_build_gen(&body->mbo_fid1);
2295 if (body->mbo_valid & OBD_MD_FLATIME) {
2296 if (body->mbo_atime > inode->i_atime.tv_sec)
2297 inode->i_atime.tv_sec = body->mbo_atime;
2298 lli->lli_atime = body->mbo_atime;
2301 if (body->mbo_valid & OBD_MD_FLMTIME) {
2302 if (body->mbo_mtime > inode->i_mtime.tv_sec) {
2304 "setting ino %lu mtime from %lld to %llu\n",
2305 inode->i_ino, (s64)inode->i_mtime.tv_sec,
2307 inode->i_mtime.tv_sec = body->mbo_mtime;
2309 lli->lli_mtime = body->mbo_mtime;
2312 if (body->mbo_valid & OBD_MD_FLCTIME) {
2313 if (body->mbo_ctime > inode->i_ctime.tv_sec)
2314 inode->i_ctime.tv_sec = body->mbo_ctime;
2315 lli->lli_ctime = body->mbo_ctime;
2318 if (body->mbo_valid & OBD_MD_FLBTIME)
2319 lli->lli_btime = body->mbo_btime;
2321 /* Clear i_flags to remove S_NOSEC before permissions are updated */
2322 if (body->mbo_valid & OBD_MD_FLFLAGS)
2323 ll_update_inode_flags(inode, body->mbo_flags);
2324 if (body->mbo_valid & OBD_MD_FLMODE)
2325 inode->i_mode = (inode->i_mode & S_IFMT) |
2326 (body->mbo_mode & ~S_IFMT);
2328 if (body->mbo_valid & OBD_MD_FLTYPE)
2329 inode->i_mode = (inode->i_mode & ~S_IFMT) |
2330 (body->mbo_mode & S_IFMT);
2332 LASSERT(inode->i_mode != 0);
2333 if (body->mbo_valid & OBD_MD_FLUID)
2334 inode->i_uid = make_kuid(&init_user_ns, body->mbo_uid);
2335 if (body->mbo_valid & OBD_MD_FLGID)
2336 inode->i_gid = make_kgid(&init_user_ns, body->mbo_gid);
2337 if (body->mbo_valid & OBD_MD_FLPROJID)
2338 lli->lli_projid = body->mbo_projid;
2339 if (body->mbo_valid & OBD_MD_FLNLINK)
2340 set_nlink(inode, body->mbo_nlink);
2341 if (body->mbo_valid & OBD_MD_FLRDEV)
2342 inode->i_rdev = old_decode_dev(body->mbo_rdev);
2344 if (body->mbo_valid & OBD_MD_FLID) {
2345 /* FID shouldn't be changed! */
2346 if (fid_is_sane(&lli->lli_fid)) {
2347 LASSERTF(lu_fid_eq(&lli->lli_fid, &body->mbo_fid1),
2348 "Trying to change FID "DFID
2349 " to the "DFID", inode "DFID"(%p)\n",
2350 PFID(&lli->lli_fid), PFID(&body->mbo_fid1),
2351 PFID(ll_inode2fid(inode)), inode);
2353 lli->lli_fid = body->mbo_fid1;
2357 LASSERT(fid_seq(&lli->lli_fid) != 0);
2359 lli->lli_attr_valid = body->mbo_valid;
2360 if (body->mbo_valid & OBD_MD_FLSIZE) {
2361 i_size_write(inode, body->mbo_size);
2363 CDEBUG(D_VFSTRACE, "inode="DFID", updating i_size %llu\n",
2364 PFID(ll_inode2fid(inode)),
2365 (unsigned long long)body->mbo_size);
2367 if (body->mbo_valid & OBD_MD_FLBLOCKS)
2368 inode->i_blocks = body->mbo_blocks;
2370 if (body->mbo_valid & OBD_MD_FLLAZYSIZE)
2371 lli->lli_lazysize = body->mbo_size;
2372 if (body->mbo_valid & OBD_MD_FLLAZYBLOCKS)
2373 lli->lli_lazyblocks = body->mbo_blocks;
2376 if (body->mbo_valid & OBD_MD_TSTATE) {
2377 /* Set LLIF_FILE_RESTORING if restore ongoing and
2378 * clear it when done to ensure to start again
2379 * glimpsing updated attrs
2381 if (body->mbo_t_state & MS_RESTORE)
2382 ll_file_set_flag(lli, LLIF_FILE_RESTORING);
2384 ll_file_clear_flag(lli, LLIF_FILE_RESTORING);
2390 int ll_read_inode2(struct inode *inode, void *opaque)
2392 struct lustre_md *md = opaque;
2393 struct ll_inode_info *lli = ll_i2info(inode);
2397 CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p)\n",
2398 PFID(&lli->lli_fid), inode);
2400 /* Core attributes from the MDS first. This is a new inode, and
2401 * the VFS doesn't zero times in the core inode so we have to do
2402 * it ourselves. They will be overwritten by either MDS or OST
2403 * attributes - we just need to make sure they aren't newer.
2405 inode->i_mtime.tv_sec = 0;
2406 inode->i_atime.tv_sec = 0;
2407 inode->i_ctime.tv_sec = 0;
2409 rc = ll_update_inode(inode, md);
2413 /* OIDEBUG(inode); */
2415 #ifdef HAVE_BACKING_DEV_INFO
2416 /* initializing backing dev info. */
2417 inode->i_mapping->backing_dev_info = &s2lsi(inode->i_sb)->lsi_bdi;
2419 if (S_ISREG(inode->i_mode)) {
2420 struct ll_sb_info *sbi = ll_i2sbi(inode);
2421 inode->i_op = &ll_file_inode_operations;
2422 inode->i_fop = sbi->ll_fop;
2423 inode->i_mapping->a_ops = (struct address_space_operations *)&ll_aops;
2425 } else if (S_ISDIR(inode->i_mode)) {
2426 inode->i_op = &ll_dir_inode_operations;
2427 inode->i_fop = &ll_dir_operations;
2429 } else if (S_ISLNK(inode->i_mode)) {
2430 inode->i_op = &ll_fast_symlink_inode_operations;
2433 inode->i_op = &ll_special_inode_operations;
2435 init_special_inode(inode, inode->i_mode,
2444 void ll_delete_inode(struct inode *inode)
2446 struct ll_inode_info *lli = ll_i2info(inode);
2447 struct address_space *mapping = &inode->i_data;
2448 unsigned long nrpages;
2449 unsigned long flags;
2453 if (S_ISREG(inode->i_mode) && lli->lli_clob != NULL) {
2454 /* It is last chance to write out dirty pages,
2455 * otherwise we may lose data while umount.
2457 * If i_nlink is 0 then just discard data. This is safe because
2458 * local inode gets i_nlink 0 from server only for the last
2459 * unlink, so that file is not opened somewhere else
2461 cl_sync_file_range(inode, 0, OBD_OBJECT_EOF, inode->i_nlink ?
2462 CL_FSYNC_LOCAL : CL_FSYNC_DISCARD, 1);
2464 truncate_inode_pages_final(mapping);
2466 /* Workaround for LU-118: Note nrpages may not be totally updated when
2467 * truncate_inode_pages() returns, as there can be a page in the process
2468 * of deletion (inside __delete_from_page_cache()) in the specified
2469 * range. Thus mapping->nrpages can be non-zero when this function
2470 * returns even after truncation of the whole mapping. Only do this if
2471 * npages isn't already zero.
2473 nrpages = mapping->nrpages;
2475 ll_xa_lock_irqsave(&mapping->i_pages, flags);
2476 nrpages = mapping->nrpages;
2477 ll_xa_unlock_irqrestore(&mapping->i_pages, flags);
2478 } /* Workaround end */
2480 LASSERTF(nrpages == 0, "%s: inode="DFID"(%p) nrpages=%lu, "
2481 "see https://jira.whamcloud.com/browse/LU-118\n",
2482 ll_i2sbi(inode)->ll_fsname,
2483 PFID(ll_inode2fid(inode)), inode, nrpages);
2485 ll_clear_inode(inode);
2491 int ll_iocontrol(struct inode *inode, struct file *file,
2492 unsigned int cmd, unsigned long arg)
2494 struct ll_sb_info *sbi = ll_i2sbi(inode);
2495 struct ptlrpc_request *req = NULL;
2500 case FS_IOC_GETFLAGS: {
2501 struct mdt_body *body;
2502 struct md_op_data *op_data;
2504 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL,
2505 0, 0, LUSTRE_OPC_ANY,
2507 if (IS_ERR(op_data))
2508 RETURN(PTR_ERR(op_data));
2510 op_data->op_valid = OBD_MD_FLFLAGS;
2511 rc = md_getattr(sbi->ll_md_exp, op_data, &req);
2512 ll_finish_md_op_data(op_data);
2514 CERROR("%s: failure inode "DFID": rc = %d\n",
2515 sbi->ll_md_exp->exp_obd->obd_name,
2516 PFID(ll_inode2fid(inode)), rc);
2520 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
2522 flags = body->mbo_flags;
2524 ptlrpc_req_finished(req);
2526 RETURN(put_user(flags, (int __user *)arg));
2528 case FS_IOC_SETFLAGS: {
2530 struct md_op_data *op_data;
2531 struct cl_object *obj;
2532 struct fsxattr fa = { 0 };
2534 if (get_user(flags, (int __user *)arg))
2537 fa.fsx_projid = ll_i2info(inode)->lli_projid;
2538 if (flags & LUSTRE_PROJINHERIT_FL)
2539 fa.fsx_xflags = FS_XFLAG_PROJINHERIT;
2541 rc = ll_ioctl_check_project(inode, &fa);
2545 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
2546 LUSTRE_OPC_ANY, NULL);
2547 if (IS_ERR(op_data))
2548 RETURN(PTR_ERR(op_data));
2550 op_data->op_attr_flags = flags;
2551 op_data->op_xvalid |= OP_XVALID_FLAGS;
2552 rc = md_setattr(sbi->ll_md_exp, op_data, NULL, 0, &req);
2553 ll_finish_md_op_data(op_data);
2554 ptlrpc_req_finished(req);
2558 ll_update_inode_flags(inode, flags);
2560 obj = ll_i2info(inode)->lli_clob;
2564 OBD_ALLOC_PTR(attr);
2568 rc = cl_setattr_ost(obj, attr, OP_XVALID_FLAGS, flags);
2580 int ll_flush_ctx(struct inode *inode)
2582 struct ll_sb_info *sbi = ll_i2sbi(inode);
2584 CDEBUG(D_SEC, "flush context for user %d\n",
2585 from_kuid(&init_user_ns, current_uid()));
2587 obd_set_info_async(NULL, sbi->ll_md_exp,
2588 sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
2590 obd_set_info_async(NULL, sbi->ll_dt_exp,
2591 sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
2596 /* umount -f client means force down, don't save state */
2597 void ll_umount_begin(struct super_block *sb)
2599 struct ll_sb_info *sbi = ll_s2sbi(sb);
2600 struct obd_device *obd;
2601 struct obd_ioctl_data *ioc_data;
2605 CDEBUG(D_VFSTRACE, "VFS Op: superblock %p count %d active %d\n", sb,
2606 sb->s_count, atomic_read(&sb->s_active));
2608 obd = class_exp2obd(sbi->ll_md_exp);
2610 CERROR("Invalid MDC connection handle %#llx\n",
2611 sbi->ll_md_exp->exp_handle.h_cookie);
2617 obd = class_exp2obd(sbi->ll_dt_exp);
2619 CERROR("Invalid LOV connection handle %#llx\n",
2620 sbi->ll_dt_exp->exp_handle.h_cookie);
2626 OBD_ALLOC_PTR(ioc_data);
2628 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_md_exp,
2629 sizeof *ioc_data, ioc_data, NULL);
2631 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_dt_exp,
2632 sizeof *ioc_data, ioc_data, NULL);
2634 OBD_FREE_PTR(ioc_data);
2637 /* Really, we'd like to wait until there are no requests outstanding,
2638 * and then continue. For now, we just periodically checking for vfs
2639 * to decrement mnt_cnt and hope to finish it within 10sec.
2643 !may_umount(sbi->ll_mnt.mnt)) {
2651 int ll_remount_fs(struct super_block *sb, int *flags, char *data)
2653 struct ll_sb_info *sbi = ll_s2sbi(sb);
2654 char *profilenm = get_profile_name(sb);
2658 if ((*flags & MS_RDONLY) != (sb->s_flags & SB_RDONLY)) {
2659 read_only = *flags & MS_RDONLY;
2660 err = obd_set_info_async(NULL, sbi->ll_md_exp,
2661 sizeof(KEY_READ_ONLY),
2662 KEY_READ_ONLY, sizeof(read_only),
2665 LCONSOLE_WARN("Failed to remount %s %s (%d)\n",
2666 profilenm, read_only ?
2667 "read-only" : "read-write", err);
2672 sb->s_flags |= SB_RDONLY;
2674 sb->s_flags &= ~SB_RDONLY;
2676 if (sbi->ll_flags & LL_SBI_VERBOSE)
2677 LCONSOLE_WARN("Remounted %s %s\n", profilenm,
2678 read_only ? "read-only" : "read-write");
2684 * Cleanup the open handle that is cached on MDT-side.
2686 * For open case, the client side open handling thread may hit error
2687 * after the MDT grant the open. Under such case, the client should
2688 * send close RPC to the MDT as cleanup; otherwise, the open handle
2689 * on the MDT will be leaked there until the client umount or evicted.
2691 * In further, if someone unlinked the file, because the open handle
2692 * holds the reference on such file/object, then it will block the
2693 * subsequent threads that want to locate such object via FID.
2695 * \param[in] sb super block for this file-system
2696 * \param[in] open_req pointer to the original open request
2698 void ll_open_cleanup(struct super_block *sb, struct ptlrpc_request *open_req)
2700 struct mdt_body *body;
2701 struct md_op_data *op_data;
2702 struct ptlrpc_request *close_req = NULL;
2703 struct obd_export *exp = ll_s2sbi(sb)->ll_md_exp;
2706 body = req_capsule_server_get(&open_req->rq_pill, &RMF_MDT_BODY);
2707 OBD_ALLOC_PTR(op_data);
2708 if (op_data == NULL) {
2709 CWARN("%s: cannot allocate op_data to release open handle for "
2710 DFID"\n", ll_s2sbi(sb)->ll_fsname, PFID(&body->mbo_fid1));
2715 op_data->op_fid1 = body->mbo_fid1;
2716 op_data->op_open_handle = body->mbo_open_handle;
2717 op_data->op_mod_time = ktime_get_real_seconds();
2718 md_close(exp, op_data, NULL, &close_req);
2719 ptlrpc_req_finished(close_req);
2720 ll_finish_md_op_data(op_data);
2725 int ll_prep_inode(struct inode **inode, struct ptlrpc_request *req,
2726 struct super_block *sb, struct lookup_intent *it)
2728 struct ll_sb_info *sbi = NULL;
2729 struct lustre_md md = { NULL };
2730 bool default_lmv_deleted = false;
2735 LASSERT(*inode || sb);
2736 sbi = sb ? ll_s2sbi(sb) : ll_i2sbi(*inode);
2737 rc = md_get_lustre_md(sbi->ll_md_exp, req, sbi->ll_dt_exp,
2738 sbi->ll_md_exp, &md);
2743 * clear default_lmv only if intent_getattr reply doesn't contain it.
2744 * but it needs to be done after iget, check this early because
2745 * ll_update_lsm_md() may change md.
2747 if (it && (it->it_op & (IT_LOOKUP | IT_GETATTR)) &&
2748 S_ISDIR(md.body->mbo_mode) && !md.default_lmv)
2749 default_lmv_deleted = true;
2752 rc = ll_update_inode(*inode, &md);
2756 LASSERT(sb != NULL);
2759 * At this point server returns to client's same fid as client
2760 * generated for creating. So using ->fid1 is okay here.
2762 if (!fid_is_sane(&md.body->mbo_fid1)) {
2763 CERROR("%s: Fid is insane "DFID"\n",
2765 PFID(&md.body->mbo_fid1));
2766 GOTO(out, rc = -EINVAL);
2769 *inode = ll_iget(sb, cl_fid_build_ino(&md.body->mbo_fid1,
2770 sbi->ll_flags & LL_SBI_32BIT_API),
2772 if (IS_ERR(*inode)) {
2774 rc = IS_ERR(*inode) ? PTR_ERR(*inode) : -ENOMEM;
2776 CERROR("new_inode -fatal: rc %d\n", rc);
2781 /* Handling piggyback layout lock.
2782 * Layout lock can be piggybacked by getattr and open request.
2783 * The lsm can be applied to inode only if it comes with a layout lock
2784 * otherwise correct layout may be overwritten, for example:
2785 * 1. proc1: mdt returns a lsm but not granting layout
2786 * 2. layout was changed by another client
2787 * 3. proc2: refresh layout and layout lock granted
2788 * 4. proc1: to apply a stale layout */
2789 if (it != NULL && it->it_lock_mode != 0) {
2790 struct lustre_handle lockh;
2791 struct ldlm_lock *lock;
2793 lockh.cookie = it->it_lock_handle;
2794 lock = ldlm_handle2lock(&lockh);
2795 LASSERT(lock != NULL);
2796 if (ldlm_has_layout(lock)) {
2797 struct cl_object_conf conf;
2799 memset(&conf, 0, sizeof(conf));
2800 conf.coc_opc = OBJECT_CONF_SET;
2801 conf.coc_inode = *inode;
2802 conf.coc_lock = lock;
2803 conf.u.coc_layout = md.layout;
2804 (void)ll_layout_conf(*inode, &conf);
2806 LDLM_LOCK_PUT(lock);
2809 if (default_lmv_deleted)
2810 ll_update_default_lsm_md(*inode, &md);
2815 /* cleanup will be done if necessary */
2816 md_free_lustre_md(sbi->ll_md_exp, &md);
2818 if (rc != 0 && it != NULL && it->it_op & IT_OPEN) {
2819 ll_intent_drop_lock(it);
2820 ll_open_cleanup(sb != NULL ? sb : (*inode)->i_sb, req);
2826 int ll_obd_statfs(struct inode *inode, void __user *arg)
2828 struct ll_sb_info *sbi = NULL;
2829 struct obd_export *exp;
2831 struct obd_ioctl_data *data = NULL;
2835 if (!inode || !(sbi = ll_i2sbi(inode)))
2836 GOTO(out_statfs, rc = -EINVAL);
2838 rc = obd_ioctl_getdata(&buf, &len, arg);
2840 GOTO(out_statfs, rc);
2843 if (!data->ioc_inlbuf1 || !data->ioc_inlbuf2 ||
2844 !data->ioc_pbuf1 || !data->ioc_pbuf2)
2845 GOTO(out_statfs, rc = -EINVAL);
2847 if (data->ioc_inllen1 != sizeof(__u32) ||
2848 data->ioc_inllen2 != sizeof(__u32) ||
2849 data->ioc_plen1 != sizeof(struct obd_statfs) ||
2850 data->ioc_plen2 != sizeof(struct obd_uuid))
2851 GOTO(out_statfs, rc = -EINVAL);
2853 memcpy(&type, data->ioc_inlbuf1, sizeof(__u32));
2854 if (type & LL_STATFS_LMV)
2855 exp = sbi->ll_md_exp;
2856 else if (type & LL_STATFS_LOV)
2857 exp = sbi->ll_dt_exp;
2859 GOTO(out_statfs, rc = -ENODEV);
2861 rc = obd_iocontrol(IOC_OBD_STATFS, exp, len, buf, NULL);
2863 GOTO(out_statfs, rc);
2865 OBD_FREE_LARGE(buf, len);
2870 * this is normally called in ll_fini_md_op_data(), but sometimes it needs to
2871 * be called early to avoid deadlock.
2873 void ll_unlock_md_op_lsm(struct md_op_data *op_data)
2875 if (op_data->op_mea2_sem) {
2876 up_read_non_owner(op_data->op_mea2_sem);
2877 op_data->op_mea2_sem = NULL;
2880 if (op_data->op_mea1_sem) {
2881 up_read_non_owner(op_data->op_mea1_sem);
2882 op_data->op_mea1_sem = NULL;
2886 /* this function prepares md_op_data hint for passing it down to MD stack. */
2887 struct md_op_data *ll_prep_md_op_data(struct md_op_data *op_data,
2888 struct inode *i1, struct inode *i2,
2889 const char *name, size_t namelen,
2890 __u32 mode, enum md_op_code opc,
2893 LASSERT(i1 != NULL);
2896 /* Do not reuse namelen for something else. */
2898 return ERR_PTR(-EINVAL);
2900 if (namelen > ll_i2sbi(i1)->ll_namelen)
2901 return ERR_PTR(-ENAMETOOLONG);
2903 if (!lu_name_is_valid_2(name, namelen))
2904 return ERR_PTR(-EINVAL);
2907 if (op_data == NULL)
2908 OBD_ALLOC_PTR(op_data);
2910 if (op_data == NULL)
2911 return ERR_PTR(-ENOMEM);
2913 ll_i2gids(op_data->op_suppgids, i1, i2);
2914 op_data->op_fid1 = *ll_inode2fid(i1);
2915 op_data->op_code = opc;
2917 if (S_ISDIR(i1->i_mode)) {
2918 down_read_non_owner(&ll_i2info(i1)->lli_lsm_sem);
2919 op_data->op_mea1_sem = &ll_i2info(i1)->lli_lsm_sem;
2920 op_data->op_mea1 = ll_i2info(i1)->lli_lsm_md;
2921 op_data->op_default_mea1 = ll_i2info(i1)->lli_default_lsm_md;
2925 op_data->op_fid2 = *ll_inode2fid(i2);
2926 if (S_ISDIR(i2->i_mode)) {
2928 /* i2 is typically a child of i1, and MUST be
2929 * further from the root to avoid deadlocks.
2931 down_read_non_owner(&ll_i2info(i2)->lli_lsm_sem);
2932 op_data->op_mea2_sem =
2933 &ll_i2info(i2)->lli_lsm_sem;
2935 op_data->op_mea2 = ll_i2info(i2)->lli_lsm_md;
2938 fid_zero(&op_data->op_fid2);
2941 if (ll_i2sbi(i1)->ll_flags & LL_SBI_64BIT_HASH)
2942 op_data->op_cli_flags |= CLI_HASH64;
2944 if (ll_need_32bit_api(ll_i2sbi(i1)))
2945 op_data->op_cli_flags |= CLI_API32;
2947 op_data->op_name = name;
2948 op_data->op_namelen = namelen;
2949 op_data->op_mode = mode;
2950 op_data->op_mod_time = ktime_get_real_seconds();
2951 op_data->op_fsuid = from_kuid(&init_user_ns, current_fsuid());
2952 op_data->op_fsgid = from_kgid(&init_user_ns, current_fsgid());
2953 op_data->op_cap = cfs_curproc_cap_pack();
2954 op_data->op_mds = 0;
2955 if ((opc == LUSTRE_OPC_CREATE) && (name != NULL) &&
2956 filename_is_volatile(name, namelen, &op_data->op_mds)) {
2957 op_data->op_bias |= MDS_CREATE_VOLATILE;
2959 op_data->op_data = data;
2964 void ll_finish_md_op_data(struct md_op_data *op_data)
2966 ll_unlock_md_op_lsm(op_data);
2967 security_release_secctx(op_data->op_file_secctx,
2968 op_data->op_file_secctx_size);
2969 llcrypt_free_ctx(op_data->op_file_encctx, op_data->op_file_encctx_size);
2970 OBD_FREE_PTR(op_data);
2973 int ll_show_options(struct seq_file *seq, struct dentry *dentry)
2975 struct ll_sb_info *sbi;
2977 LASSERT(seq && dentry);
2978 sbi = ll_s2sbi(dentry->d_sb);
2980 if (sbi->ll_flags & LL_SBI_NOLCK)
2981 seq_puts(seq, ",nolock");
2983 /* "flock" is the default since 2.13, but it wasn't for many years,
2984 * so it is still useful to print this to show it is enabled.
2985 * Start to print "noflock" so it is now clear when flock is disabled.
2987 if (sbi->ll_flags & LL_SBI_FLOCK)
2988 seq_puts(seq, ",flock");
2989 else if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
2990 seq_puts(seq, ",localflock");
2992 seq_puts(seq, ",noflock");
2994 if (sbi->ll_flags & LL_SBI_USER_XATTR)
2995 seq_puts(seq, ",user_xattr");
2997 if (sbi->ll_flags & LL_SBI_LAZYSTATFS)
2998 seq_puts(seq, ",lazystatfs");
3000 if (sbi->ll_flags & LL_SBI_USER_FID2PATH)
3001 seq_puts(seq, ",user_fid2path");
3003 if (sbi->ll_flags & LL_SBI_ALWAYS_PING)
3004 seq_puts(seq, ",always_ping");
3006 if (ll_sbi_has_test_dummy_encryption(sbi))
3007 seq_puts(seq, ",test_dummy_encryption");
3009 if (ll_sbi_has_encrypt(sbi))
3010 seq_puts(seq, ",encrypt");
3012 seq_puts(seq, ",noencrypt");
3018 * Get obd name by cmd, and copy out to user space
3020 int ll_get_obd_name(struct inode *inode, unsigned int cmd, unsigned long arg)
3022 struct ll_sb_info *sbi = ll_i2sbi(inode);
3023 struct obd_device *obd;
3026 if (cmd == OBD_IOC_GETDTNAME)
3027 obd = class_exp2obd(sbi->ll_dt_exp);
3028 else if (cmd == OBD_IOC_GETMDNAME)
3029 obd = class_exp2obd(sbi->ll_md_exp);
3036 if (copy_to_user((void __user *)arg, obd->obd_name,
3037 strlen(obd->obd_name) + 1))
3043 static char* ll_d_path(struct dentry *dentry, char *buf, int bufsize)
3050 p.mnt = current->fs->root.mnt;
3052 path = d_path(&p, buf, bufsize);
3057 void ll_dirty_page_discard_warn(struct page *page, int ioret)
3059 char *buf, *path = NULL;
3060 struct dentry *dentry = NULL;
3061 struct inode *inode = page->mapping->host;
3063 /* this can be called inside spin lock so use GFP_ATOMIC. */
3064 buf = (char *)__get_free_page(GFP_ATOMIC);
3066 dentry = d_find_alias(page->mapping->host);
3068 path = ll_d_path(dentry, buf, PAGE_SIZE);
3071 /* The below message is checked in recovery-small.sh test_24b */
3073 "%s: dirty page discard: %s/fid: "DFID"/%s may get corrupted "
3074 "(rc %d)\n", ll_i2sbi(inode)->ll_fsname,
3075 s2lsi(page->mapping->host->i_sb)->lsi_lmd->lmd_dev,
3076 PFID(ll_inode2fid(inode)),
3077 (path && !IS_ERR(path)) ? path : "", ioret);
3083 free_page((unsigned long)buf);
3086 ssize_t ll_copy_user_md(const struct lov_user_md __user *md,
3087 struct lov_user_md **kbuf)
3089 struct lov_user_md lum;
3093 if (copy_from_user(&lum, md, sizeof(lum)))
3096 lum_size = ll_lov_user_md_size(&lum);
3100 OBD_ALLOC_LARGE(*kbuf, lum_size);
3104 if (copy_from_user(*kbuf, md, lum_size) != 0) {
3105 OBD_FREE_LARGE(*kbuf, lum_size);
3113 * Compute llite root squash state after a change of root squash
3114 * configuration setting or add/remove of a lnet nid
3116 void ll_compute_rootsquash_state(struct ll_sb_info *sbi)
3118 struct root_squash_info *squash = &sbi->ll_squash;
3121 struct lnet_process_id id;
3123 /* Update norootsquash flag */
3124 spin_lock(&squash->rsi_lock);
3125 if (list_empty(&squash->rsi_nosquash_nids))
3126 sbi->ll_flags &= ~LL_SBI_NOROOTSQUASH;
3128 /* Do not apply root squash as soon as one of our NIDs is
3129 * in the nosquash_nids list */
3132 while (LNetGetId(i++, &id) != -ENOENT) {
3133 if (id.nid == LNET_NID_LO_0)
3135 if (cfs_match_nid(id.nid, &squash->rsi_nosquash_nids)) {
3141 sbi->ll_flags |= LL_SBI_NOROOTSQUASH;
3143 sbi->ll_flags &= ~LL_SBI_NOROOTSQUASH;
3145 spin_unlock(&squash->rsi_lock);
3149 * Parse linkea content to extract information about a given hardlink
3151 * \param[in] ldata - Initialized linkea data
3152 * \param[in] linkno - Link identifier
3153 * \param[out] parent_fid - The entry's parent FID
3154 * \param[out] ln - Entry name destination buffer
3156 * \retval 0 on success
3157 * \retval Appropriate negative error code on failure
3159 static int ll_linkea_decode(struct linkea_data *ldata, unsigned int linkno,
3160 struct lu_fid *parent_fid, struct lu_name *ln)
3166 rc = linkea_init_with_rec(ldata);
3170 if (linkno >= ldata->ld_leh->leh_reccount)
3171 /* beyond last link */
3174 linkea_first_entry(ldata);
3175 for (idx = 0; ldata->ld_lee != NULL; idx++) {
3176 linkea_entry_unpack(ldata->ld_lee, &ldata->ld_reclen, ln,
3181 linkea_next_entry(ldata);
3191 * Get parent FID and name of an identified link. Operation is performed for
3192 * a given link number, letting the caller iterate over linkno to list one or
3193 * all links of an entry.
3195 * \param[in] file - File descriptor against which to perform the operation
3196 * \param[in,out] arg - User-filled structure containing the linkno to operate
3197 * on and the available size. It is eventually filled with
3198 * the requested information or left untouched on error
3200 * \retval - 0 on success
3201 * \retval - Appropriate negative error code on failure
3203 int ll_getparent(struct file *file, struct getparent __user *arg)
3205 struct inode *inode = file_inode(file);
3206 struct linkea_data *ldata;
3207 struct lu_buf buf = LU_BUF_NULL;
3209 struct lu_fid parent_fid;
3216 if (!cfs_capable(CFS_CAP_DAC_READ_SEARCH) &&
3217 !(ll_i2sbi(inode)->ll_flags & LL_SBI_USER_FID2PATH))
3220 if (get_user(name_size, &arg->gp_name_size))
3223 if (get_user(linkno, &arg->gp_linkno))
3226 if (name_size > PATH_MAX)
3229 OBD_ALLOC(ldata, sizeof(*ldata));
3233 rc = linkea_data_new(ldata, &buf);
3235 GOTO(ldata_free, rc);
3237 rc = ll_xattr_list(inode, XATTR_NAME_LINK, XATTR_TRUSTED_T, buf.lb_buf,
3238 buf.lb_len, OBD_MD_FLXATTR);
3242 rc = ll_linkea_decode(ldata, linkno, &parent_fid, &ln);
3246 if (ln.ln_namelen >= name_size)
3247 GOTO(lb_free, rc = -EOVERFLOW);
3249 if (copy_to_user(&arg->gp_fid, &parent_fid, sizeof(arg->gp_fid)))
3250 GOTO(lb_free, rc = -EFAULT);
3252 if (copy_to_user(&arg->gp_name, ln.ln_name, ln.ln_namelen))
3253 GOTO(lb_free, rc = -EFAULT);
3255 if (put_user('\0', arg->gp_name + ln.ln_namelen))
3256 GOTO(lb_free, rc = -EFAULT);
3261 OBD_FREE(ldata, sizeof(*ldata));