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 |
280 OBD_CONNECT2_GETATTR_PFID;
282 #ifdef HAVE_LRU_RESIZE_SUPPORT
283 if (sbi->ll_flags & LL_SBI_LRU_RESIZE)
284 data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
286 data->ocd_connect_flags |= OBD_CONNECT_ACL_FLAGS;
288 data->ocd_cksum_types = obd_cksum_types_supported_client();
290 if (OBD_FAIL_CHECK(OBD_FAIL_MDC_LIGHTWEIGHT))
291 /* flag mdc connection as lightweight, only used for test
292 * purpose, use with care */
293 data->ocd_connect_flags |= OBD_CONNECT_LIGHTWEIGHT;
295 data->ocd_ibits_known = MDS_INODELOCK_FULL;
296 data->ocd_version = LUSTRE_VERSION_CODE;
298 if (sb->s_flags & SB_RDONLY)
299 data->ocd_connect_flags |= OBD_CONNECT_RDONLY;
300 if (sbi->ll_flags & LL_SBI_USER_XATTR)
301 data->ocd_connect_flags |= OBD_CONNECT_XATTR;
304 /* Setting this indicates we correctly support S_NOSEC (See kernel
305 * commit 9e1f1de02c2275d7172e18dc4e7c2065777611bf)
307 sb->s_flags |= SB_NOSEC;
310 if (sbi->ll_flags & LL_SBI_FLOCK)
311 sbi->ll_fop = &ll_file_operations_flock;
312 else if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
313 sbi->ll_fop = &ll_file_operations;
315 sbi->ll_fop = &ll_file_operations_noflock;
317 /* always ping even if server suppress_pings */
318 if (sbi->ll_flags & LL_SBI_ALWAYS_PING)
319 data->ocd_connect_flags &= ~OBD_CONNECT_PINGLESS;
321 obd_connect_set_secctx(data);
322 if (ll_sbi_has_encrypt(sbi))
323 obd_connect_set_enc(data);
325 #if defined(CONFIG_SECURITY)
326 data->ocd_connect_flags2 |= OBD_CONNECT2_SELINUX_POLICY;
329 data->ocd_brw_size = MD_MAX_BRW_SIZE;
331 err = obd_connect(NULL, &sbi->ll_md_exp, sbi->ll_md_obd,
332 &sbi->ll_sb_uuid, data, sbi->ll_cache);
334 LCONSOLE_ERROR_MSG(0x14f, "An MDT (md %s) is performing "
335 "recovery, of which this client is not a "
336 "part. Please wait for recovery to complete,"
337 " abort, or time out.\n", md);
340 CERROR("cannot connect to %s: rc = %d\n", md, err);
344 sbi->ll_md_exp->exp_connect_data = *data;
346 err = obd_fid_init(sbi->ll_md_exp->exp_obd, sbi->ll_md_exp,
347 LUSTRE_SEQ_METADATA);
349 CERROR("%s: Can't init metadata layer FID infrastructure, "
350 "rc = %d\n", sbi->ll_md_exp->exp_obd->obd_name, err);
354 /* For mount, we only need fs info from MDT0, and also in DNE, it
355 * can make sure the client can be mounted as long as MDT0 is
357 err = obd_statfs(NULL, sbi->ll_md_exp, osfs,
358 ktime_get_seconds() - sbi->ll_statfs_max_age,
359 OBD_STATFS_FOR_MDT0);
361 GOTO(out_md_fid, err);
363 /* This needs to be after statfs to ensure connect has finished.
364 * Note that "data" does NOT contain the valid connect reply.
365 * If connecting to a 1.8 server there will be no LMV device, so
366 * we can access the MDC export directly and exp_connect_flags will
367 * be non-zero, but if accessing an upgraded 2.1 server it will
368 * have the correct flags filled in.
369 * XXX: fill in the LMV exp_connect_flags from MDC(s). */
370 valid = exp_connect_flags(sbi->ll_md_exp) & CLIENT_CONNECT_MDT_REQD;
371 if (exp_connect_flags(sbi->ll_md_exp) != 0 &&
372 valid != CLIENT_CONNECT_MDT_REQD) {
375 OBD_ALLOC_WAIT(buf, PAGE_SIZE);
376 obd_connect_flags2str(buf, PAGE_SIZE,
377 valid ^ CLIENT_CONNECT_MDT_REQD, 0, ",");
378 LCONSOLE_ERROR_MSG(0x170, "Server %s does not support "
379 "feature(s) needed for correct operation "
380 "of this client (%s). Please upgrade "
381 "server or downgrade client.\n",
382 sbi->ll_md_exp->exp_obd->obd_name, buf);
383 OBD_FREE(buf, PAGE_SIZE);
384 GOTO(out_md_fid, err = -EPROTO);
387 size = sizeof(*data);
388 err = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_CONN_DATA),
389 KEY_CONN_DATA, &size, data);
391 CERROR("%s: Get connect data failed: rc = %d\n",
392 sbi->ll_md_exp->exp_obd->obd_name, err);
393 GOTO(out_md_fid, err);
396 LASSERT(osfs->os_bsize);
397 sb->s_blocksize = osfs->os_bsize;
398 sb->s_blocksize_bits = log2(osfs->os_bsize);
399 sb->s_magic = LL_SUPER_MAGIC;
400 sb->s_maxbytes = MAX_LFS_FILESIZE;
401 sbi->ll_namelen = osfs->os_namelen;
402 sbi->ll_mnt.mnt = current->fs->root.mnt;
404 if ((sbi->ll_flags & LL_SBI_USER_XATTR) &&
405 !(data->ocd_connect_flags & OBD_CONNECT_XATTR)) {
406 LCONSOLE_INFO("Disabling user_xattr feature because "
407 "it is not supported on the server\n");
408 sbi->ll_flags &= ~LL_SBI_USER_XATTR;
411 if (data->ocd_connect_flags & OBD_CONNECT_ACL) {
413 sb->s_flags |= SB_POSIXACL;
415 sbi->ll_flags |= LL_SBI_ACL;
417 LCONSOLE_INFO("client wants to enable acl, but mdt not!\n");
419 sb->s_flags &= ~SB_POSIXACL;
421 sbi->ll_flags &= ~LL_SBI_ACL;
424 if (data->ocd_connect_flags & OBD_CONNECT_64BITHASH)
425 sbi->ll_flags |= LL_SBI_64BIT_HASH;
427 if (data->ocd_connect_flags & OBD_CONNECT_LAYOUTLOCK)
428 sbi->ll_flags |= LL_SBI_LAYOUT_LOCK;
430 if (obd_connect_has_secctx(data))
431 sbi->ll_flags |= LL_SBI_FILE_SECCTX;
433 if (ll_sbi_has_encrypt(sbi) && !obd_connect_has_enc(data)) {
434 if (ll_sbi_has_test_dummy_encryption(sbi))
435 LCONSOLE_WARN("%s: server %s does not support encryption feature, encryption deactivated.\n",
437 sbi->ll_md_exp->exp_obd->obd_name);
438 ll_sbi_set_encrypt(sbi, false);
441 if (data->ocd_ibits_known & MDS_INODELOCK_XATTR) {
442 if (!(data->ocd_connect_flags & OBD_CONNECT_MAX_EASIZE)) {
443 LCONSOLE_INFO("%s: disabling xattr cache due to "
444 "unknown maximum xattr size.\n", dt);
445 } else if (!sbi->ll_xattr_cache_set) {
446 /* If xattr_cache is already set (no matter 0 or 1)
447 * during processing llog, it won't be enabled here. */
448 sbi->ll_flags |= LL_SBI_XATTR_CACHE;
449 sbi->ll_xattr_cache_enabled = 1;
453 sbi->ll_dt_obd = class_name2obd(dt);
454 if (!sbi->ll_dt_obd) {
455 CERROR("DT %s: not setup or attached\n", dt);
456 GOTO(out_md_fid, err = -ENODEV);
459 /* pass client page size via ocd_grant_blkbits, the server should report
460 * back its backend blocksize for grant calculation purpose */
461 data->ocd_grant_blkbits = PAGE_SHIFT;
463 /* indicate OST features supported by this client */
464 data->ocd_connect_flags = OBD_CONNECT_GRANT | OBD_CONNECT_VERSION |
465 OBD_CONNECT_REQPORTAL | OBD_CONNECT_BRW_SIZE |
466 OBD_CONNECT_CANCELSET | OBD_CONNECT_FID |
467 OBD_CONNECT_SRVLOCK | OBD_CONNECT_TRUNCLOCK|
468 OBD_CONNECT_AT | OBD_CONNECT_OSS_CAPA |
469 OBD_CONNECT_VBR | OBD_CONNECT_FULL20 |
470 OBD_CONNECT_64BITHASH | OBD_CONNECT_MAXBYTES |
471 OBD_CONNECT_EINPROGRESS |
472 OBD_CONNECT_JOBSTATS | OBD_CONNECT_LVB_TYPE |
473 OBD_CONNECT_LAYOUTLOCK |
474 OBD_CONNECT_PINGLESS | OBD_CONNECT_LFSCK |
475 OBD_CONNECT_BULK_MBITS | OBD_CONNECT_SHORTIO |
476 OBD_CONNECT_FLAGS2 | OBD_CONNECT_GRANT_SHRINK;
477 data->ocd_connect_flags2 = OBD_CONNECT2_LOCKAHEAD |
478 OBD_CONNECT2_INC_XID;
480 if (!OBD_FAIL_CHECK(OBD_FAIL_OSC_CONNECT_GRANT_PARAM))
481 data->ocd_connect_flags |= OBD_CONNECT_GRANT_PARAM;
483 /* OBD_CONNECT_CKSUM should always be set, even if checksums are
484 * disabled by default, because it can still be enabled on the
485 * fly via /sys. As a consequence, we still need to come to an
486 * agreement on the supported algorithms at connect time
488 data->ocd_connect_flags |= OBD_CONNECT_CKSUM;
490 if (OBD_FAIL_CHECK(OBD_FAIL_OSC_CKSUM_ADLER_ONLY))
491 data->ocd_cksum_types = OBD_CKSUM_ADLER;
493 data->ocd_cksum_types = obd_cksum_types_supported_client();
495 #ifdef HAVE_LRU_RESIZE_SUPPORT
496 data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
498 /* always ping even if server suppress_pings */
499 if (sbi->ll_flags & LL_SBI_ALWAYS_PING)
500 data->ocd_connect_flags &= ~OBD_CONNECT_PINGLESS;
502 if (ll_sbi_has_encrypt(sbi))
503 obd_connect_set_enc(data);
505 CDEBUG(D_RPCTRACE, "ocd_connect_flags: %#llx ocd_version: %d "
506 "ocd_grant: %d\n", data->ocd_connect_flags,
507 data->ocd_version, data->ocd_grant);
509 sbi->ll_dt_obd->obd_upcall.onu_owner = &sbi->ll_lco;
510 sbi->ll_dt_obd->obd_upcall.onu_upcall = cl_ocd_update;
512 data->ocd_brw_size = DT_MAX_BRW_SIZE;
514 err = obd_connect(NULL, &sbi->ll_dt_exp, sbi->ll_dt_obd,
515 &sbi->ll_sb_uuid, data, sbi->ll_cache);
517 LCONSOLE_ERROR_MSG(0x150, "An OST (dt %s) is performing "
518 "recovery, of which this client is not a "
519 "part. Please wait for recovery to "
520 "complete, abort, or time out.\n", dt);
523 CERROR("%s: Cannot connect to %s: rc = %d\n",
524 sbi->ll_dt_exp->exp_obd->obd_name, dt, err);
528 if (ll_sbi_has_encrypt(sbi) &&
529 !obd_connect_has_enc(&sbi->ll_dt_obd->u.lov.lov_ocd)) {
530 if (ll_sbi_has_test_dummy_encryption(sbi))
531 LCONSOLE_WARN("%s: server %s does not support encryption feature, encryption deactivated.\n",
533 ll_sbi_set_encrypt(sbi, false);
534 } else if (ll_sbi_has_test_dummy_encryption(sbi)) {
535 LCONSOLE_WARN("Test dummy encryption mode enabled\n");
538 sbi->ll_dt_exp->exp_connect_data = *data;
540 /* Don't change value if it was specified in the config log */
541 if (sbi->ll_ra_info.ra_max_read_ahead_whole_pages == -1) {
542 sbi->ll_ra_info.ra_max_read_ahead_whole_pages =
543 max_t(unsigned long, SBI_DEFAULT_READ_AHEAD_WHOLE_MAX,
544 (data->ocd_brw_size >> PAGE_SHIFT));
545 if (sbi->ll_ra_info.ra_max_read_ahead_whole_pages >
546 sbi->ll_ra_info.ra_max_pages_per_file)
547 sbi->ll_ra_info.ra_max_read_ahead_whole_pages =
548 sbi->ll_ra_info.ra_max_pages_per_file;
551 err = obd_fid_init(sbi->ll_dt_exp->exp_obd, sbi->ll_dt_exp,
552 LUSTRE_SEQ_METADATA);
554 CERROR("%s: Can't init data layer FID infrastructure, "
555 "rc = %d\n", sbi->ll_dt_exp->exp_obd->obd_name, err);
559 mutex_lock(&sbi->ll_lco.lco_lock);
560 sbi->ll_lco.lco_flags = data->ocd_connect_flags;
561 sbi->ll_lco.lco_md_exp = sbi->ll_md_exp;
562 sbi->ll_lco.lco_dt_exp = sbi->ll_dt_exp;
563 mutex_unlock(&sbi->ll_lco.lco_lock);
565 fid_zero(&sbi->ll_root_fid);
566 err = md_get_root(sbi->ll_md_exp, get_mount_fileset(sb),
569 CERROR("cannot mds_connect: rc = %d\n", err);
570 GOTO(out_lock_cn_cb, err);
572 if (!fid_is_sane(&sbi->ll_root_fid)) {
573 CERROR("%s: Invalid root fid "DFID" during mount\n",
574 sbi->ll_md_exp->exp_obd->obd_name,
575 PFID(&sbi->ll_root_fid));
576 GOTO(out_lock_cn_cb, err = -EINVAL);
578 CDEBUG(D_SUPER, "rootfid "DFID"\n", PFID(&sbi->ll_root_fid));
580 sb->s_op = &lustre_super_operations;
581 sb->s_xattr = ll_xattr_handlers;
582 #if THREAD_SIZE >= 8192 /*b=17630*/
583 sb->s_export_op = &lustre_export_operations;
585 #ifdef HAVE_LUSTRE_CRYPTO
586 llcrypt_set_ops(sb, &lustre_cryptops);
590 * XXX: move this to after cbd setup? */
591 valid = OBD_MD_FLGETATTR | OBD_MD_FLBLOCKS | OBD_MD_FLMODEASIZE;
592 if (sbi->ll_flags & LL_SBI_ACL)
593 valid |= OBD_MD_FLACL;
595 OBD_ALLOC_PTR(op_data);
597 GOTO(out_lock_cn_cb, err = -ENOMEM);
599 op_data->op_fid1 = sbi->ll_root_fid;
600 op_data->op_mode = 0;
601 op_data->op_valid = valid;
603 err = md_getattr(sbi->ll_md_exp, op_data, &request);
605 OBD_FREE_PTR(op_data);
607 CERROR("%s: md_getattr failed for root: rc = %d\n",
608 sbi->ll_md_exp->exp_obd->obd_name, err);
609 GOTO(out_lock_cn_cb, err);
612 err = md_get_lustre_md(sbi->ll_md_exp, request, sbi->ll_dt_exp,
613 sbi->ll_md_exp, &lmd);
615 CERROR("failed to understand root inode md: rc = %d\n", err);
616 ptlrpc_req_finished(request);
617 GOTO(out_lock_cn_cb, err);
620 LASSERT(fid_is_sane(&sbi->ll_root_fid));
621 root = ll_iget(sb, cl_fid_build_ino(&sbi->ll_root_fid,
622 sbi->ll_flags & LL_SBI_32BIT_API),
624 md_free_lustre_md(sbi->ll_md_exp, &lmd);
625 ptlrpc_req_finished(request);
629 err = IS_ERR(root) ? PTR_ERR(root) : -EBADF;
631 CERROR("%s: bad ll_iget() for root: rc = %d\n",
632 sbi->ll_fsname, err);
636 checksum = sbi->ll_flags & LL_SBI_CHECKSUM;
637 if (sbi->ll_checksum_set) {
638 err = obd_set_info_async(NULL, sbi->ll_dt_exp,
639 sizeof(KEY_CHECKSUM), KEY_CHECKSUM,
640 sizeof(checksum), &checksum, NULL);
642 CERROR("%s: Set checksum failed: rc = %d\n",
643 sbi->ll_dt_exp->exp_obd->obd_name, err);
649 sb->s_root = d_make_root(root);
650 if (sb->s_root == NULL) {
652 CERROR("%s: can't make root dentry: rc = %d\n",
653 sbi->ll_fsname, err);
657 sbi->ll_sdev_orig = sb->s_dev;
659 /* We set sb->s_dev equal on all lustre clients in order to support
660 * NFS export clustering. NFSD requires that the FSID be the same
662 /* s_dev is also used in lt_compare() to compare two fs, but that is
663 * only a node-local comparison. */
664 uuid = obd_get_uuid(sbi->ll_md_exp);
666 sb->s_dev = get_uuid2int(uuid->uuid, strlen(uuid->uuid));
673 if (sbi->ll_dt_obd) {
674 err = sysfs_create_link(&sbi->ll_kset.kobj,
675 &sbi->ll_dt_obd->obd_kset.kobj,
676 sbi->ll_dt_obd->obd_type->typ_name);
678 CERROR("%s: could not register %s in llite: rc = %d\n",
679 dt, sbi->ll_fsname, err);
684 if (sbi->ll_md_obd) {
685 err = sysfs_create_link(&sbi->ll_kset.kobj,
686 &sbi->ll_md_obd->obd_kset.kobj,
687 sbi->ll_md_obd->obd_type->typ_name);
689 CERROR("%s: could not register %s in llite: rc = %d\n",
690 md, sbi->ll_fsname, err);
700 obd_fid_fini(sbi->ll_dt_exp->exp_obd);
702 obd_disconnect(sbi->ll_dt_exp);
703 sbi->ll_dt_exp = NULL;
704 sbi->ll_dt_obd = NULL;
706 obd_fid_fini(sbi->ll_md_exp->exp_obd);
708 obd_disconnect(sbi->ll_md_exp);
709 sbi->ll_md_exp = NULL;
710 sbi->ll_md_obd = NULL;
719 int ll_get_max_mdsize(struct ll_sb_info *sbi, int *lmmsize)
723 size = sizeof(*lmmsize);
724 rc = obd_get_info(NULL, sbi->ll_dt_exp, sizeof(KEY_MAX_EASIZE),
725 KEY_MAX_EASIZE, &size, lmmsize);
727 CERROR("%s: cannot get max LOV EA size: rc = %d\n",
728 sbi->ll_dt_exp->exp_obd->obd_name, rc);
732 CDEBUG(D_INFO, "max LOV ea size: %d\n", *lmmsize);
735 rc = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_MAX_EASIZE),
736 KEY_MAX_EASIZE, &size, lmmsize);
738 CERROR("Get max mdsize error rc %d\n", rc);
740 CDEBUG(D_INFO, "max LMV ea size: %d\n", *lmmsize);
746 * Get the value of the default_easize parameter.
748 * \see client_obd::cl_default_mds_easize
750 * \param[in] sbi superblock info for this filesystem
751 * \param[out] lmmsize pointer to storage location for value
753 * \retval 0 on success
754 * \retval negative negated errno on failure
756 int ll_get_default_mdsize(struct ll_sb_info *sbi, int *lmmsize)
761 rc = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_DEFAULT_EASIZE),
762 KEY_DEFAULT_EASIZE, &size, lmmsize);
764 CERROR("Get default mdsize error rc %d\n", rc);
770 * Set the default_easize parameter to the given value.
772 * \see client_obd::cl_default_mds_easize
774 * \param[in] sbi superblock info for this filesystem
775 * \param[in] lmmsize the size to set
777 * \retval 0 on success
778 * \retval negative negated errno on failure
780 int ll_set_default_mdsize(struct ll_sb_info *sbi, int lmmsize)
784 if (lmmsize < sizeof(struct lov_mds_md) ||
785 lmmsize > OBD_MAX_DEFAULT_EA_SIZE)
788 rc = obd_set_info_async(NULL, sbi->ll_md_exp,
789 sizeof(KEY_DEFAULT_EASIZE), KEY_DEFAULT_EASIZE,
790 sizeof(int), &lmmsize, NULL);
795 static void client_common_put_super(struct super_block *sb)
797 struct ll_sb_info *sbi = ll_s2sbi(sb);
802 obd_fid_fini(sbi->ll_dt_exp->exp_obd);
803 obd_disconnect(sbi->ll_dt_exp);
804 sbi->ll_dt_exp = NULL;
806 ll_debugfs_unregister_super(sb);
808 obd_fid_fini(sbi->ll_md_exp->exp_obd);
809 obd_disconnect(sbi->ll_md_exp);
810 sbi->ll_md_exp = NULL;
815 void ll_kill_super(struct super_block *sb)
817 struct ll_sb_info *sbi;
821 if (!(sb->s_flags & SB_ACTIVE))
825 /* we need restore s_dev from changed for clustred NFS before put_super
826 * because new kernels have cached s_dev and change sb->s_dev in
827 * put_super not affected real removing devices */
829 sb->s_dev = sbi->ll_sdev_orig;
831 /* wait running statahead threads to quit */
832 while (atomic_read(&sbi->ll_sa_running) > 0)
833 schedule_timeout_uninterruptible(
834 cfs_time_seconds(1) >> 3);
840 static inline int ll_set_opt(const char *opt, char *data, int fl)
842 if (strncmp(opt, data, strlen(opt)) != 0)
848 /* non-client-specific mount options are parsed in lmd_parse */
849 static int ll_options(char *options, struct ll_sb_info *sbi)
852 char *s1 = options, *s2;
853 int *flags = &sbi->ll_flags;
859 CDEBUG(D_CONFIG, "Parsing opts %s\n", options);
862 CDEBUG(D_SUPER, "next opt=%s\n", s1);
863 tmp = ll_set_opt("nolock", s1, LL_SBI_NOLCK);
868 tmp = ll_set_opt("flock", s1, LL_SBI_FLOCK);
870 *flags = (*flags & ~LL_SBI_LOCALFLOCK) | tmp;
873 tmp = ll_set_opt("localflock", s1, LL_SBI_LOCALFLOCK);
875 *flags = (*flags & ~LL_SBI_FLOCK) | tmp;
878 tmp = ll_set_opt("noflock", s1, LL_SBI_FLOCK|LL_SBI_LOCALFLOCK);
883 tmp = ll_set_opt("user_xattr", s1, LL_SBI_USER_XATTR);
888 tmp = ll_set_opt("nouser_xattr", s1, LL_SBI_USER_XATTR);
893 tmp = ll_set_opt("context", s1, 1);
896 tmp = ll_set_opt("fscontext", s1, 1);
899 tmp = ll_set_opt("defcontext", s1, 1);
902 tmp = ll_set_opt("rootcontext", s1, 1);
905 tmp = ll_set_opt("user_fid2path", s1, LL_SBI_USER_FID2PATH);
910 tmp = ll_set_opt("nouser_fid2path", s1, LL_SBI_USER_FID2PATH);
916 tmp = ll_set_opt("checksum", s1, LL_SBI_CHECKSUM);
919 sbi->ll_checksum_set = 1;
922 tmp = ll_set_opt("nochecksum", s1, LL_SBI_CHECKSUM);
925 sbi->ll_checksum_set = 1;
928 tmp = ll_set_opt("lruresize", s1, LL_SBI_LRU_RESIZE);
933 tmp = ll_set_opt("nolruresize", s1, LL_SBI_LRU_RESIZE);
938 tmp = ll_set_opt("lazystatfs", s1, LL_SBI_LAZYSTATFS);
943 tmp = ll_set_opt("nolazystatfs", s1, LL_SBI_LAZYSTATFS);
948 tmp = ll_set_opt("32bitapi", s1, LL_SBI_32BIT_API);
953 tmp = ll_set_opt("verbose", s1, LL_SBI_VERBOSE);
958 tmp = ll_set_opt("noverbose", s1, LL_SBI_VERBOSE);
963 tmp = ll_set_opt("always_ping", s1, LL_SBI_ALWAYS_PING);
968 tmp = ll_set_opt("test_dummy_encryption", s1,
969 LL_SBI_TEST_DUMMY_ENCRYPTION);
971 #ifdef HAVE_LUSTRE_CRYPTO
974 LCONSOLE_WARN("Test dummy encryption mount option ignored: encryption not supported\n");
978 tmp = ll_set_opt("noencrypt", s1, LL_SBI_ENCRYPT);
980 #ifdef HAVE_LUSTRE_CRYPTO
983 LCONSOLE_WARN("noencrypt mount option ignored: encryption not supported\n");
987 LCONSOLE_ERROR_MSG(0x152, "Unknown option '%s', won't mount.\n",
993 s2 = strchr(s1, ',');
1001 void ll_lli_init(struct ll_inode_info *lli)
1003 lli->lli_inode_magic = LLI_INODE_MAGIC;
1005 spin_lock_init(&lli->lli_lock);
1006 lli->lli_posix_acl = NULL;
1007 /* Do not set lli_fid, it has been initialized already. */
1008 fid_zero(&lli->lli_pfid);
1009 lli->lli_mds_read_och = NULL;
1010 lli->lli_mds_write_och = NULL;
1011 lli->lli_mds_exec_och = NULL;
1012 lli->lli_open_fd_read_count = 0;
1013 lli->lli_open_fd_write_count = 0;
1014 lli->lli_open_fd_exec_count = 0;
1015 mutex_init(&lli->lli_och_mutex);
1016 spin_lock_init(&lli->lli_agl_lock);
1017 spin_lock_init(&lli->lli_layout_lock);
1018 ll_layout_version_set(lli, CL_LAYOUT_GEN_NONE);
1019 lli->lli_clob = NULL;
1021 init_rwsem(&lli->lli_xattrs_list_rwsem);
1022 mutex_init(&lli->lli_xattrs_enq_lock);
1024 LASSERT(lli->lli_vfs_inode.i_mode != 0);
1025 if (S_ISDIR(lli->lli_vfs_inode.i_mode)) {
1026 lli->lli_opendir_key = NULL;
1027 lli->lli_sai = NULL;
1028 spin_lock_init(&lli->lli_sa_lock);
1029 lli->lli_opendir_pid = 0;
1030 lli->lli_sa_enabled = 0;
1031 init_rwsem(&lli->lli_lsm_sem);
1033 mutex_init(&lli->lli_size_mutex);
1034 mutex_init(&lli->lli_setattr_mutex);
1035 lli->lli_symlink_name = NULL;
1036 ll_trunc_sem_init(&lli->lli_trunc_sem);
1037 range_lock_tree_init(&lli->lli_write_tree);
1038 init_rwsem(&lli->lli_glimpse_sem);
1039 lli->lli_glimpse_time = ktime_set(0, 0);
1040 INIT_LIST_HEAD(&lli->lli_agl_list);
1041 lli->lli_agl_index = 0;
1042 lli->lli_async_rc = 0;
1043 spin_lock_init(&lli->lli_heat_lock);
1044 obd_heat_clear(lli->lli_heat_instances, OBD_HEAT_COUNT);
1045 lli->lli_heat_flags = 0;
1046 mutex_init(&lli->lli_pcc_lock);
1047 lli->lli_pcc_state = PCC_STATE_FL_NONE;
1048 lli->lli_pcc_inode = NULL;
1049 lli->lli_pcc_dsflags = PCC_DATASET_INVALID;
1050 lli->lli_pcc_generation = 0;
1051 mutex_init(&lli->lli_group_mutex);
1052 lli->lli_group_users = 0;
1053 lli->lli_group_gid = 0;
1055 mutex_init(&lli->lli_layout_mutex);
1056 memset(lli->lli_jobid, 0, sizeof(lli->lli_jobid));
1059 #define MAX_STRING_SIZE 128
1061 #ifndef HAVE_SUPER_SETUP_BDI_NAME
1063 #define LSI_BDI_INITIALIZED 0x00400000
1065 #ifndef HAVE_BDI_CAP_MAP_COPY
1066 # define BDI_CAP_MAP_COPY 0
1069 static int super_setup_bdi_name(struct super_block *sb, char *fmt, ...)
1071 struct lustre_sb_info *lsi = s2lsi(sb);
1072 char buf[MAX_STRING_SIZE];
1076 err = bdi_init(&lsi->lsi_bdi);
1080 lsi->lsi_flags |= LSI_BDI_INITIALIZED;
1081 lsi->lsi_bdi.capabilities = BDI_CAP_MAP_COPY;
1082 lsi->lsi_bdi.name = "lustre";
1083 va_start(args, fmt);
1084 vsnprintf(buf, MAX_STRING_SIZE, fmt, args);
1086 err = bdi_register(&lsi->lsi_bdi, NULL, "%s", buf);
1089 sb->s_bdi = &lsi->lsi_bdi;
1093 #endif /* !HAVE_SUPER_SETUP_BDI_NAME */
1095 int ll_fill_super(struct super_block *sb)
1097 struct lustre_profile *lprof = NULL;
1098 struct lustre_sb_info *lsi = s2lsi(sb);
1099 struct ll_sb_info *sbi = NULL;
1100 char *dt = NULL, *md = NULL;
1101 char *profilenm = get_profile_name(sb);
1102 struct config_llog_instance *cfg;
1103 /* %p for void* in printf needs 16+2 characters: 0xffffffffffffffff */
1104 const int instlen = LUSTRE_MAXINSTANCE + 2;
1105 unsigned long cfg_instance = ll_get_cfg_instance(sb);
1106 char name[MAX_STRING_SIZE];
1115 /* for ASLR, to map between cfg_instance and hashed ptr */
1116 CDEBUG(D_VFSTRACE, "VFS Op: cfg_instance %s-%016lx (sb %p)\n",
1117 profilenm, cfg_instance, sb);
1121 GOTO(out_free_cfg, err = -ENOMEM);
1123 /* client additional sb info */
1124 lsi->lsi_llsbi = sbi = ll_init_sbi();
1126 GOTO(out_free_cfg, err = PTR_ERR(sbi));
1128 err = ll_options(lsi->lsi_lmd->lmd_opts, sbi);
1130 GOTO(out_free_cfg, err);
1132 /* kernel >= 2.6.38 store dentry operations in sb->s_d_op. */
1133 sb->s_d_op = &ll_d_ops;
1136 generate_random_uuid(uuid.b);
1137 snprintf(sbi->ll_sb_uuid.uuid, sizeof(sbi->ll_sb_uuid), "%pU", uuid.b);
1139 CDEBUG(D_CONFIG, "llite sb uuid: %s\n", sbi->ll_sb_uuid.uuid);
1142 len = strlen(profilenm);
1143 ptr = strrchr(profilenm, '-');
1144 if (ptr && (strcmp(ptr, "-client") == 0))
1147 if (len > LUSTRE_MAXFSNAME) {
1148 if (unlikely(len >= MAX_STRING_SIZE))
1149 len = MAX_STRING_SIZE - 1;
1150 strncpy(name, profilenm, len);
1152 err = -ENAMETOOLONG;
1153 CERROR("%s: fsname longer than %u characters: rc = %d\n",
1154 name, LUSTRE_MAXFSNAME, err);
1155 GOTO(out_free_cfg, err);
1157 strncpy(sbi->ll_fsname, profilenm, len);
1158 sbi->ll_fsname[len] = '\0';
1161 snprintf(name, sizeof(name), "%.*s-%016lx", len,
1162 profilenm, cfg_instance);
1164 err = super_setup_bdi_name(sb, "%s", name);
1166 GOTO(out_free_cfg, err);
1168 /* Call ll_debugfs_register_super() before lustre_process_log()
1169 * so that "llite.*.*" params can be processed correctly.
1171 err = ll_debugfs_register_super(sb, name);
1173 CERROR("%s: could not register mountpoint in llite: rc = %d\n",
1174 sbi->ll_fsname, err);
1178 /* The cfg_instance is a value unique to this super, in case some
1179 * joker tries to mount the same fs at two mount points.
1181 cfg->cfg_instance = cfg_instance;
1182 cfg->cfg_uuid = lsi->lsi_llsbi->ll_sb_uuid;
1183 cfg->cfg_callback = class_config_llog_handler;
1184 cfg->cfg_sub_clds = CONFIG_SUB_CLIENT;
1185 /* set up client obds */
1186 err = lustre_process_log(sb, profilenm, cfg);
1188 GOTO(out_debugfs, err);
1190 /* Profile set with LCFG_MOUNTOPT so we can find our mdc and osc obds */
1191 lprof = class_get_profile(profilenm);
1192 if (lprof == NULL) {
1193 LCONSOLE_ERROR_MSG(0x156, "The client profile '%s' could not be"
1194 " read from the MGS. Does that filesystem "
1195 "exist?\n", profilenm);
1196 GOTO(out_debugfs, err = -EINVAL);
1198 CDEBUG(D_CONFIG, "Found profile %s: mdc=%s osc=%s\n", profilenm,
1199 lprof->lp_md, lprof->lp_dt);
1201 dt_len = strlen(lprof->lp_dt) + instlen + 2;
1202 OBD_ALLOC(dt, dt_len);
1204 GOTO(out_profile, err = -ENOMEM);
1205 snprintf(dt, dt_len - 1, "%s-%016lx", lprof->lp_dt, cfg_instance);
1207 md_len = strlen(lprof->lp_md) + instlen + 2;
1208 OBD_ALLOC(md, md_len);
1210 GOTO(out_free_dt, err = -ENOMEM);
1211 snprintf(md, md_len - 1, "%s-%016lx", lprof->lp_md, cfg_instance);
1213 /* connections, registrations, sb setup */
1214 err = client_common_fill_super(sb, md, dt);
1216 GOTO(out_free_md, err);
1218 sbi->ll_client_common_fill_super_succeeded = 1;
1222 OBD_FREE(md, md_len);
1225 OBD_FREE(dt, dt_len);
1228 class_put_profile(lprof);
1231 ll_debugfs_unregister_super(sb);
1238 else if (sbi->ll_flags & LL_SBI_VERBOSE)
1239 LCONSOLE_WARN("Mounted %s\n", profilenm);
1241 } /* ll_fill_super */
1243 void ll_put_super(struct super_block *sb)
1245 struct config_llog_instance cfg, params_cfg;
1246 struct obd_device *obd;
1247 struct lustre_sb_info *lsi = s2lsi(sb);
1248 struct ll_sb_info *sbi = ll_s2sbi(sb);
1249 char *profilenm = get_profile_name(sb);
1250 unsigned long cfg_instance = ll_get_cfg_instance(sb);
1252 int next, force = 1, rc = 0;
1256 GOTO(out_no_sbi, 0);
1258 /* Should replace instance_id with something better for ASLR */
1259 CDEBUG(D_VFSTRACE, "VFS Op: cfg_instance %s-%016lx (sb %p)\n",
1260 profilenm, cfg_instance, sb);
1262 cfg.cfg_instance = cfg_instance;
1263 lustre_end_log(sb, profilenm, &cfg);
1265 params_cfg.cfg_instance = cfg_instance;
1266 lustre_end_log(sb, PARAMS_FILENAME, ¶ms_cfg);
1268 if (sbi->ll_md_exp) {
1269 obd = class_exp2obd(sbi->ll_md_exp);
1271 force = obd->obd_force;
1274 /* Wait for unstable pages to be committed to stable storage */
1276 rc = l_wait_event_abortable(
1277 sbi->ll_cache->ccc_unstable_waitq,
1278 atomic_long_read(&sbi->ll_cache->ccc_unstable_nr) == 0);
1281 ccc_count = atomic_long_read(&sbi->ll_cache->ccc_unstable_nr);
1282 if (force == 0 && rc != -ERESTARTSYS)
1283 LASSERTF(ccc_count == 0, "count: %li\n", ccc_count);
1285 /* We need to set force before the lov_disconnect in
1286 * lustre_common_put_super, since l_d cleans up osc's as well.
1290 while ((obd = class_devices_in_group(&sbi->ll_sb_uuid,
1292 obd->obd_force = force;
1296 if (sbi->ll_client_common_fill_super_succeeded) {
1297 /* Only if client_common_fill_super succeeded */
1298 client_common_put_super(sb);
1302 while ((obd = class_devices_in_group(&sbi->ll_sb_uuid, &next)))
1303 class_manual_cleanup(obd);
1305 if (sbi->ll_flags & LL_SBI_VERBOSE)
1306 LCONSOLE_WARN("Unmounted %s\n", profilenm ? profilenm : "");
1309 class_del_profile(profilenm);
1311 #ifndef HAVE_SUPER_SETUP_BDI_NAME
1312 if (lsi->lsi_flags & LSI_BDI_INITIALIZED) {
1313 bdi_destroy(&lsi->lsi_bdi);
1314 lsi->lsi_flags &= ~LSI_BDI_INITIALIZED;
1319 lsi->lsi_llsbi = NULL;
1321 lustre_common_put_super(sb);
1323 cl_env_cache_purge(~0);
1325 module_put(THIS_MODULE);
1328 } /* client_put_super */
1330 struct inode *ll_inode_from_resource_lock(struct ldlm_lock *lock)
1332 struct inode *inode = NULL;
1334 /* NOTE: we depend on atomic igrab() -bzzz */
1335 lock_res_and_lock(lock);
1336 if (lock->l_resource->lr_lvb_inode) {
1337 struct ll_inode_info * lli;
1338 lli = ll_i2info(lock->l_resource->lr_lvb_inode);
1339 if (lli->lli_inode_magic == LLI_INODE_MAGIC) {
1340 inode = igrab(lock->l_resource->lr_lvb_inode);
1342 inode = lock->l_resource->lr_lvb_inode;
1343 LDLM_DEBUG_LIMIT(inode->i_state & I_FREEING ? D_INFO :
1344 D_WARNING, lock, "lr_lvb_inode %p is "
1345 "bogus: magic %08x",
1346 lock->l_resource->lr_lvb_inode,
1347 lli->lli_inode_magic);
1351 unlock_res_and_lock(lock);
1355 void ll_dir_clear_lsm_md(struct inode *inode)
1357 struct ll_inode_info *lli = ll_i2info(inode);
1359 LASSERT(S_ISDIR(inode->i_mode));
1361 if (lli->lli_lsm_md) {
1362 lmv_free_memmd(lli->lli_lsm_md);
1363 lli->lli_lsm_md = NULL;
1366 if (lli->lli_default_lsm_md) {
1367 lmv_free_memmd(lli->lli_default_lsm_md);
1368 lli->lli_default_lsm_md = NULL;
1372 static struct inode *ll_iget_anon_dir(struct super_block *sb,
1373 const struct lu_fid *fid,
1374 struct lustre_md *md)
1376 struct ll_sb_info *sbi = ll_s2sbi(sb);
1377 struct mdt_body *body = md->body;
1378 struct inode *inode;
1382 ino = cl_fid_build_ino(fid, sbi->ll_flags & LL_SBI_32BIT_API);
1383 inode = iget_locked(sb, ino);
1384 if (inode == NULL) {
1385 CERROR("%s: failed get simple inode "DFID": rc = -ENOENT\n",
1386 sbi->ll_fsname, PFID(fid));
1387 RETURN(ERR_PTR(-ENOENT));
1390 if (inode->i_state & I_NEW) {
1391 struct ll_inode_info *lli = ll_i2info(inode);
1392 struct lmv_stripe_md *lsm = md->lmv;
1394 inode->i_mode = (inode->i_mode & ~S_IFMT) |
1395 (body->mbo_mode & S_IFMT);
1396 LASSERTF(S_ISDIR(inode->i_mode), "Not slave inode "DFID"\n",
1399 inode->i_mtime.tv_sec = 0;
1400 inode->i_atime.tv_sec = 0;
1401 inode->i_ctime.tv_sec = 0;
1404 #ifdef HAVE_BACKING_DEV_INFO
1405 /* initializing backing dev info. */
1406 inode->i_mapping->backing_dev_info =
1407 &s2lsi(inode->i_sb)->lsi_bdi;
1409 inode->i_op = &ll_dir_inode_operations;
1410 inode->i_fop = &ll_dir_operations;
1411 lli->lli_fid = *fid;
1414 LASSERT(lsm != NULL);
1415 /* master object FID */
1416 lli->lli_pfid = body->mbo_fid1;
1417 CDEBUG(D_INODE, "lli %p slave "DFID" master "DFID"\n",
1418 lli, PFID(fid), PFID(&lli->lli_pfid));
1419 unlock_new_inode(inode);
1425 static int ll_init_lsm_md(struct inode *inode, struct lustre_md *md)
1428 struct lmv_stripe_md *lsm = md->lmv;
1429 struct ll_inode_info *lli = ll_i2info(inode);
1432 LASSERT(lsm != NULL);
1434 CDEBUG(D_INODE, "%s: "DFID" set dir layout:\n",
1435 ll_i2sbi(inode)->ll_fsname, PFID(&lli->lli_fid));
1436 lsm_md_dump(D_INODE, lsm);
1438 if (!lmv_dir_striped(lsm))
1441 /* XXX sigh, this lsm_root initialization should be in
1442 * LMV layer, but it needs ll_iget right now, so we
1443 * put this here right now. */
1444 for (i = 0; i < lsm->lsm_md_stripe_count; i++) {
1445 fid = &lsm->lsm_md_oinfo[i].lmo_fid;
1446 LASSERT(lsm->lsm_md_oinfo[i].lmo_root == NULL);
1448 if (!fid_is_sane(fid))
1451 /* Unfortunately ll_iget will call ll_update_inode,
1452 * where the initialization of slave inode is slightly
1453 * different, so it reset lsm_md to NULL to avoid
1454 * initializing lsm for slave inode. */
1455 lsm->lsm_md_oinfo[i].lmo_root =
1456 ll_iget_anon_dir(inode->i_sb, fid, md);
1457 if (IS_ERR(lsm->lsm_md_oinfo[i].lmo_root)) {
1458 int rc = PTR_ERR(lsm->lsm_md_oinfo[i].lmo_root);
1460 lsm->lsm_md_oinfo[i].lmo_root = NULL;
1462 iput(lsm->lsm_md_oinfo[i].lmo_root);
1463 lsm->lsm_md_oinfo[i].lmo_root = NULL;
1469 lli->lli_lsm_md = lsm;
1474 static void ll_update_default_lsm_md(struct inode *inode, struct lustre_md *md)
1476 struct ll_inode_info *lli = ll_i2info(inode);
1478 if (!md->default_lmv) {
1479 /* clear default lsm */
1480 if (lli->lli_default_lsm_md) {
1481 down_write(&lli->lli_lsm_sem);
1482 if (lli->lli_default_lsm_md) {
1483 lmv_free_memmd(lli->lli_default_lsm_md);
1484 lli->lli_default_lsm_md = NULL;
1486 up_write(&lli->lli_lsm_sem);
1488 } else if (lli->lli_default_lsm_md) {
1489 /* update default lsm if it changes */
1490 down_read(&lli->lli_lsm_sem);
1491 if (lli->lli_default_lsm_md &&
1492 !lsm_md_eq(lli->lli_default_lsm_md, md->default_lmv)) {
1493 up_read(&lli->lli_lsm_sem);
1494 down_write(&lli->lli_lsm_sem);
1495 if (lli->lli_default_lsm_md)
1496 lmv_free_memmd(lli->lli_default_lsm_md);
1497 lli->lli_default_lsm_md = md->default_lmv;
1498 lsm_md_dump(D_INODE, md->default_lmv);
1499 md->default_lmv = NULL;
1500 up_write(&lli->lli_lsm_sem);
1502 up_read(&lli->lli_lsm_sem);
1505 /* init default lsm */
1506 down_write(&lli->lli_lsm_sem);
1507 lli->lli_default_lsm_md = md->default_lmv;
1508 lsm_md_dump(D_INODE, md->default_lmv);
1509 md->default_lmv = NULL;
1510 up_write(&lli->lli_lsm_sem);
1514 static int ll_update_lsm_md(struct inode *inode, struct lustre_md *md)
1516 struct ll_inode_info *lli = ll_i2info(inode);
1517 struct lmv_stripe_md *lsm = md->lmv;
1518 struct cl_attr *attr;
1523 LASSERT(S_ISDIR(inode->i_mode));
1524 CDEBUG(D_INODE, "update lsm %p of "DFID"\n", lli->lli_lsm_md,
1525 PFID(ll_inode2fid(inode)));
1527 /* update default LMV */
1528 if (md->default_lmv)
1529 ll_update_default_lsm_md(inode, md);
1532 * no striped information from request, lustre_md from req does not
1533 * include stripeEA, see ll_md_setattr()
1539 * normally dir layout doesn't change, only take read lock to check
1540 * that to avoid blocking other MD operations.
1542 down_read(&lli->lli_lsm_sem);
1544 /* some current lookup initialized lsm, and unchanged */
1545 if (lli->lli_lsm_md && lsm_md_eq(lli->lli_lsm_md, lsm))
1546 GOTO(unlock, rc = 0);
1548 /* if dir layout doesn't match, check whether version is increased,
1549 * which means layout is changed, this happens in dir split/merge and
1552 * foreign LMV should not change.
1554 if (lli->lli_lsm_md && lmv_dir_striped(lli->lli_lsm_md) &&
1555 lsm->lsm_md_layout_version <=
1556 lli->lli_lsm_md->lsm_md_layout_version) {
1557 CERROR("%s: "DFID" dir layout mismatch:\n",
1558 ll_i2sbi(inode)->ll_fsname, PFID(&lli->lli_fid));
1559 lsm_md_dump(D_ERROR, lli->lli_lsm_md);
1560 lsm_md_dump(D_ERROR, lsm);
1561 GOTO(unlock, rc = -EINVAL);
1564 up_read(&lli->lli_lsm_sem);
1565 down_write(&lli->lli_lsm_sem);
1566 /* clear existing lsm */
1567 if (lli->lli_lsm_md) {
1568 lmv_free_memmd(lli->lli_lsm_md);
1569 lli->lli_lsm_md = NULL;
1572 rc = ll_init_lsm_md(inode, md);
1573 up_write(&lli->lli_lsm_sem);
1578 /* set md->lmv to NULL, so the following free lustre_md will not free
1583 /* md_merge_attr() may take long, since lsm is already set, switch to
1586 down_read(&lli->lli_lsm_sem);
1588 if (!lmv_dir_striped(lli->lli_lsm_md))
1589 GOTO(unlock, rc = 0);
1591 OBD_ALLOC_PTR(attr);
1593 GOTO(unlock, rc = -ENOMEM);
1595 /* validate the lsm */
1596 rc = md_merge_attr(ll_i2mdexp(inode), &lli->lli_fid, lli->lli_lsm_md,
1597 attr, ll_md_blocking_ast);
1599 if (md->body->mbo_valid & OBD_MD_FLNLINK)
1600 md->body->mbo_nlink = attr->cat_nlink;
1601 if (md->body->mbo_valid & OBD_MD_FLSIZE)
1602 md->body->mbo_size = attr->cat_size;
1603 if (md->body->mbo_valid & OBD_MD_FLATIME)
1604 md->body->mbo_atime = attr->cat_atime;
1605 if (md->body->mbo_valid & OBD_MD_FLCTIME)
1606 md->body->mbo_ctime = attr->cat_ctime;
1607 if (md->body->mbo_valid & OBD_MD_FLMTIME)
1608 md->body->mbo_mtime = attr->cat_mtime;
1614 up_read(&lli->lli_lsm_sem);
1619 void ll_clear_inode(struct inode *inode)
1621 struct ll_inode_info *lli = ll_i2info(inode);
1622 struct ll_sb_info *sbi = ll_i2sbi(inode);
1626 CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p)\n",
1627 PFID(ll_inode2fid(inode)), inode);
1629 if (S_ISDIR(inode->i_mode)) {
1630 /* these should have been cleared in ll_file_release */
1631 LASSERT(lli->lli_opendir_key == NULL);
1632 LASSERT(lli->lli_sai == NULL);
1633 LASSERT(lli->lli_opendir_pid == 0);
1635 pcc_inode_free(inode);
1638 md_null_inode(sbi->ll_md_exp, ll_inode2fid(inode));
1640 LASSERT(!lli->lli_open_fd_write_count);
1641 LASSERT(!lli->lli_open_fd_read_count);
1642 LASSERT(!lli->lli_open_fd_exec_count);
1644 if (lli->lli_mds_write_och)
1645 ll_md_real_close(inode, FMODE_WRITE);
1646 if (lli->lli_mds_exec_och)
1647 ll_md_real_close(inode, FMODE_EXEC);
1648 if (lli->lli_mds_read_och)
1649 ll_md_real_close(inode, FMODE_READ);
1651 if (S_ISLNK(inode->i_mode) && lli->lli_symlink_name) {
1652 OBD_FREE(lli->lli_symlink_name,
1653 strlen(lli->lli_symlink_name) + 1);
1654 lli->lli_symlink_name = NULL;
1657 ll_xattr_cache_destroy(inode);
1659 forget_all_cached_acls(inode);
1661 lli->lli_inode_magic = LLI_INODE_DEAD;
1663 if (S_ISDIR(inode->i_mode))
1664 ll_dir_clear_lsm_md(inode);
1665 else if (S_ISREG(inode->i_mode) && !is_bad_inode(inode))
1666 LASSERT(list_empty(&lli->lli_agl_list));
1669 * XXX This has to be done before lsm is freed below, because
1670 * cl_object still uses inode lsm.
1672 cl_inode_fini(inode);
1674 llcrypt_put_encryption_info(inode);
1679 static int ll_md_setattr(struct dentry *dentry, struct md_op_data *op_data)
1681 struct lustre_md md;
1682 struct inode *inode = dentry->d_inode;
1683 struct ll_sb_info *sbi = ll_i2sbi(inode);
1684 struct ptlrpc_request *request = NULL;
1688 op_data = ll_prep_md_op_data(op_data, inode, NULL, NULL, 0, 0,
1689 LUSTRE_OPC_ANY, NULL);
1690 if (IS_ERR(op_data))
1691 RETURN(PTR_ERR(op_data));
1693 rc = md_setattr(sbi->ll_md_exp, op_data, NULL, 0, &request);
1695 ptlrpc_req_finished(request);
1696 if (rc == -ENOENT) {
1698 /* Unlinked special device node? Or just a race?
1699 * Pretend we done everything. */
1700 if (!S_ISREG(inode->i_mode) &&
1701 !S_ISDIR(inode->i_mode)) {
1702 ia_valid = op_data->op_attr.ia_valid;
1703 op_data->op_attr.ia_valid &= ~TIMES_SET_FLAGS;
1704 rc = simple_setattr(dentry, &op_data->op_attr);
1705 op_data->op_attr.ia_valid = ia_valid;
1707 } else if (rc != -EPERM && rc != -EACCES && rc != -ETXTBSY) {
1708 CERROR("md_setattr fails: rc = %d\n", rc);
1713 rc = md_get_lustre_md(sbi->ll_md_exp, request, sbi->ll_dt_exp,
1714 sbi->ll_md_exp, &md);
1716 ptlrpc_req_finished(request);
1720 ia_valid = op_data->op_attr.ia_valid;
1721 /* inode size will be in ll_setattr_ost, can't do it now since dirty
1722 * cache is not cleared yet. */
1723 op_data->op_attr.ia_valid &= ~(TIMES_SET_FLAGS | ATTR_SIZE);
1724 if (S_ISREG(inode->i_mode))
1726 rc = simple_setattr(dentry, &op_data->op_attr);
1727 if (S_ISREG(inode->i_mode))
1728 inode_unlock(inode);
1729 op_data->op_attr.ia_valid = ia_valid;
1731 rc = ll_update_inode(inode, &md);
1732 ptlrpc_req_finished(request);
1738 * Zero portion of page that is part of @inode.
1739 * This implies, if necessary:
1740 * - taking cl_lock on range corresponding to concerned page
1741 * - grabbing vm page
1742 * - associating cl_page
1743 * - proceeding to clio read
1744 * - zeroing range in page
1745 * - proceeding to cl_page flush
1746 * - releasing cl_lock
1748 * \param[in] inode inode
1749 * \param[in] index page index
1750 * \param[in] offset offset in page to start zero from
1751 * \param[in] len len to zero
1753 * \retval 0 on success
1754 * \retval negative errno on failure
1756 int ll_io_zero_page(struct inode *inode, pgoff_t index, pgoff_t offset,
1759 struct ll_inode_info *lli = ll_i2info(inode);
1760 struct cl_object *clob = lli->lli_clob;
1762 struct lu_env *env = NULL;
1763 struct cl_io *io = NULL;
1764 struct cl_page *clpage = NULL;
1765 struct page *vmpage = NULL;
1766 unsigned from = index << PAGE_SHIFT;
1767 struct cl_lock *lock = NULL;
1768 struct cl_lock_descr *descr = NULL;
1769 struct cl_2queue *queue = NULL;
1770 struct cl_sync_io *anchor = NULL;
1771 bool holdinglock = false;
1772 bool lockedbymyself = true;
1777 env = cl_env_get(&refcheck);
1779 RETURN(PTR_ERR(env));
1781 io = vvp_env_thread_io(env);
1783 rc = cl_io_rw_init(env, io, CIT_WRITE, from, PAGE_SIZE);
1787 lock = vvp_env_lock(env);
1788 descr = &lock->cll_descr;
1789 descr->cld_obj = io->ci_obj;
1790 descr->cld_start = cl_index(io->ci_obj, from);
1791 descr->cld_end = cl_index(io->ci_obj, from + PAGE_SIZE - 1);
1792 descr->cld_mode = CLM_WRITE;
1793 descr->cld_enq_flags = CEF_MUST | CEF_NONBLOCK;
1795 /* request lock for page */
1796 rc = cl_lock_request(env, io, lock);
1797 /* -ECANCELED indicates a matching lock with a different extent
1798 * was already present, and -EEXIST indicates a matching lock
1799 * on exactly the same extent was already present.
1800 * In both cases it means we are covered.
1802 if (rc == -ECANCELED || rc == -EEXIST)
1810 vmpage = grab_cache_page_nowait(inode->i_mapping, index);
1812 GOTO(rellock, rc = -EOPNOTSUPP);
1814 if (!PageDirty(vmpage)) {
1815 /* associate cl_page */
1816 clpage = cl_page_find(env, clob, vmpage->index,
1817 vmpage, CPT_CACHEABLE);
1819 GOTO(pagefini, rc = PTR_ERR(clpage));
1821 cl_page_assume(env, io, clpage);
1824 if (!PageUptodate(vmpage) && !PageDirty(vmpage) &&
1825 !PageWriteback(vmpage)) {
1827 /* set PagePrivate2 to detect special case of empty page
1828 * in osc_brw_fini_request()
1830 SetPagePrivate2(vmpage);
1831 rc = ll_io_read_page(env, io, clpage, NULL);
1832 if (!PagePrivate2(vmpage))
1833 /* PagePrivate2 was cleared in osc_brw_fini_request()
1834 * meaning we read an empty page. In this case, in order
1835 * to avoid allocating unnecessary block in truncated
1836 * file, we must not zero and write as below. Subsequent
1837 * server-side truncate will handle things correctly.
1839 GOTO(clpfini, rc = 0);
1840 ClearPagePrivate2(vmpage);
1843 lockedbymyself = trylock_page(vmpage);
1844 cl_page_assume(env, io, clpage);
1847 /* zero range in page */
1848 zero_user(vmpage, offset, len);
1850 if (holdinglock && clpage) {
1851 /* explicitly write newly modified page */
1852 queue = &io->ci_queue;
1853 cl_2queue_init(queue);
1854 anchor = &vvp_env_info(env)->vti_anchor;
1855 cl_sync_io_init(anchor, 1);
1856 clpage->cp_sync_io = anchor;
1857 cl_2queue_add(queue, clpage);
1858 rc = cl_io_submit_rw(env, io, CRT_WRITE, queue);
1860 GOTO(queuefini1, rc);
1861 rc = cl_sync_io_wait(env, anchor, 0);
1863 GOTO(queuefini2, rc);
1864 cl_page_assume(env, io, clpage);
1867 cl_2queue_discard(env, io, queue);
1869 cl_2queue_disown(env, io, queue);
1870 cl_2queue_fini(env, queue);
1875 cl_page_put(env, clpage);
1877 if (lockedbymyself) {
1878 unlock_page(vmpage);
1883 cl_lock_release(env, lock);
1885 cl_io_fini(env, io);
1888 cl_env_put(env, &refcheck);
1893 /* If this inode has objects allocated to it (lsm != NULL), then the OST
1894 * object(s) determine the file size and mtime. Otherwise, the MDS will
1895 * keep these values until such a time that objects are allocated for it.
1896 * We do the MDS operations first, as it is checking permissions for us.
1897 * We don't to the MDS RPC if there is nothing that we want to store there,
1898 * otherwise there is no harm in updating mtime/atime on the MDS if we are
1899 * going to do an RPC anyways.
1901 * If we are doing a truncate, we will send the mtime and ctime updates
1902 * to the OST with the punch RPC, otherwise we do an explicit setattr RPC.
1903 * I don't believe it is possible to get e.g. ATTR_MTIME_SET and ATTR_SIZE
1906 * In case of HSMimport, we only set attr on MDS.
1908 int ll_setattr_raw(struct dentry *dentry, struct iattr *attr,
1909 enum op_xvalid xvalid, bool hsm_import)
1911 struct inode *inode = dentry->d_inode;
1912 struct ll_inode_info *lli = ll_i2info(inode);
1913 struct md_op_data *op_data = NULL;
1914 ktime_t kstart = ktime_get();
1919 CDEBUG(D_VFSTRACE, "%s: setattr inode "DFID"(%p) from %llu to %llu, "
1920 "valid %x, hsm_import %d\n",
1921 ll_i2sbi(inode)->ll_fsname, PFID(&lli->lli_fid),
1922 inode, i_size_read(inode), attr->ia_size, attr->ia_valid,
1925 if (attr->ia_valid & ATTR_SIZE) {
1926 /* Check new size against VFS/VM file size limit and rlimit */
1927 rc = inode_newsize_ok(inode, attr->ia_size);
1931 /* The maximum Lustre file size is variable, based on the
1932 * OST maximum object size and number of stripes. This
1933 * needs another check in addition to the VFS check above. */
1934 if (attr->ia_size > ll_file_maxbytes(inode)) {
1935 CDEBUG(D_INODE,"file "DFID" too large %llu > %llu\n",
1936 PFID(&lli->lli_fid), attr->ia_size,
1937 ll_file_maxbytes(inode));
1941 attr->ia_valid |= ATTR_MTIME | ATTR_CTIME;
1944 /* POSIX: check before ATTR_*TIME_SET set (from inode_change_ok) */
1945 if (attr->ia_valid & TIMES_SET_FLAGS) {
1946 if ((!uid_eq(current_fsuid(), inode->i_uid)) &&
1947 !cfs_capable(CFS_CAP_FOWNER))
1951 /* We mark all of the fields "set" so MDS/OST does not re-set them */
1952 if (!(xvalid & OP_XVALID_CTIME_SET) &&
1953 (attr->ia_valid & ATTR_CTIME)) {
1954 attr->ia_ctime = current_time(inode);
1955 xvalid |= OP_XVALID_CTIME_SET;
1957 if (!(attr->ia_valid & ATTR_ATIME_SET) &&
1958 (attr->ia_valid & ATTR_ATIME)) {
1959 attr->ia_atime = current_time(inode);
1960 attr->ia_valid |= ATTR_ATIME_SET;
1962 if (!(attr->ia_valid & ATTR_MTIME_SET) &&
1963 (attr->ia_valid & ATTR_MTIME)) {
1964 attr->ia_mtime = current_time(inode);
1965 attr->ia_valid |= ATTR_MTIME_SET;
1968 if (attr->ia_valid & (ATTR_MTIME | ATTR_CTIME))
1969 CDEBUG(D_INODE, "setting mtime %lld, ctime %lld, now = %lld\n",
1970 (s64)attr->ia_mtime.tv_sec, (s64)attr->ia_ctime.tv_sec,
1971 ktime_get_real_seconds());
1973 if (S_ISREG(inode->i_mode))
1974 inode_unlock(inode);
1976 /* We always do an MDS RPC, even if we're only changing the size;
1977 * only the MDS knows whether truncate() should fail with -ETXTBUSY */
1979 OBD_ALLOC_PTR(op_data);
1980 if (op_data == NULL)
1981 GOTO(out, rc = -ENOMEM);
1983 if (!hsm_import && attr->ia_valid & ATTR_SIZE) {
1984 /* If we are changing file size, file content is
1985 * modified, flag it.
1987 xvalid |= OP_XVALID_OWNEROVERRIDE;
1988 op_data->op_bias |= MDS_DATA_MODIFIED;
1989 ll_file_clear_flag(lli, LLIF_DATA_MODIFIED);
1992 if (attr->ia_valid & ATTR_FILE) {
1993 struct ll_file_data *fd = attr->ia_file->private_data;
1995 if (fd->fd_lease_och)
1996 op_data->op_bias |= MDS_TRUNC_KEEP_LEASE;
1999 op_data->op_attr = *attr;
2000 op_data->op_xvalid = xvalid;
2002 rc = ll_md_setattr(dentry, op_data);
2006 if (!S_ISREG(inode->i_mode) || hsm_import)
2009 if (attr->ia_valid & (ATTR_SIZE | ATTR_ATIME | ATTR_ATIME_SET |
2010 ATTR_MTIME | ATTR_MTIME_SET | ATTR_CTIME) ||
2011 xvalid & OP_XVALID_CTIME_SET) {
2012 bool cached = false;
2014 rc = pcc_inode_setattr(inode, attr, &cached);
2017 CERROR("%s: PCC inode "DFID" setattr failed: "
2019 ll_i2sbi(inode)->ll_fsname,
2020 PFID(&lli->lli_fid), rc);
2024 unsigned int flags = 0;
2026 /* For truncate and utimes sending attributes to OSTs,
2027 * setting mtime/atime to the past will be performed
2028 * under PW [0:EOF] extent lock (new_size:EOF for
2029 * truncate). It may seem excessive to send mtime/atime
2030 * updates to OSTs when not setting times to past, but
2031 * it is necessary due to possible time
2032 * de-synchronization between MDT inode and OST objects
2034 if (S_ISREG(inode->i_mode) && IS_ENCRYPTED(inode) &&
2035 attr->ia_valid & ATTR_SIZE) {
2036 xvalid |= OP_XVALID_FLAGS;
2037 flags = LUSTRE_ENCRYPT_FL;
2038 /* Call to ll_io_zero_page is not necessary if
2039 * truncating on PAGE_SIZE boundary, because
2040 * whole pages will be wiped.
2041 * In case of Direct IO, all we need is to set
2044 if (attr->ia_size & ~PAGE_MASK &&
2045 !(attr->ia_valid & ATTR_FILE &&
2046 attr->ia_file->f_flags & O_DIRECT)) {
2048 attr->ia_size & (PAGE_SIZE - 1);
2050 rc = ll_io_zero_page(inode,
2051 attr->ia_size >> PAGE_SHIFT,
2052 offset, PAGE_SIZE - offset);
2057 rc = cl_setattr_ost(lli->lli_clob, attr, xvalid, flags);
2061 /* If the file was restored, it needs to set dirty flag.
2063 * We've already sent MDS_DATA_MODIFIED flag in
2064 * ll_md_setattr() for truncate. However, the MDT refuses to
2065 * set the HS_DIRTY flag on released files, so we have to set
2066 * it again if the file has been restored. Please check how
2067 * LLIF_DATA_MODIFIED is set in vvp_io_setattr_fini().
2069 * Please notice that if the file is not released, the previous
2070 * MDS_DATA_MODIFIED has taken effect and usually
2071 * LLIF_DATA_MODIFIED is not set(see vvp_io_setattr_fini()).
2072 * This way we can save an RPC for common open + trunc
2074 if (ll_file_test_and_clear_flag(lli, LLIF_DATA_MODIFIED)) {
2075 struct hsm_state_set hss = {
2076 .hss_valid = HSS_SETMASK,
2077 .hss_setmask = HS_DIRTY,
2081 rc2 = ll_hsm_state_set(inode, &hss);
2082 /* truncate and write can happen at the same time, so that
2083 * the file can be set modified even though the file is not
2084 * restored from released state, and ll_hsm_state_set() is
2085 * not applicable for the file, and rc2 < 0 is normal in this
2088 CDEBUG(D_INFO, DFID "HSM set dirty failed: rc2 = %d\n",
2089 PFID(ll_inode2fid(inode)), rc2);
2094 if (op_data != NULL)
2095 ll_finish_md_op_data(op_data);
2097 if (S_ISREG(inode->i_mode)) {
2099 if ((attr->ia_valid & ATTR_SIZE) && !hsm_import)
2100 inode_dio_wait(inode);
2101 /* Once we've got the i_mutex, it's safe to set the S_NOSEC
2102 * flag. ll_update_inode (called from ll_md_setattr), clears
2103 * inode flags, so there is a gap where S_NOSEC is not set.
2104 * This can cause a writer to take the i_mutex unnecessarily,
2105 * but this is safe to do and should be rare. */
2106 inode_has_no_xattr(inode);
2110 ll_stats_ops_tally(ll_i2sbi(inode), attr->ia_valid & ATTR_SIZE ?
2111 LPROC_LL_TRUNC : LPROC_LL_SETATTR,
2112 ktime_us_delta(ktime_get(), kstart));
2117 int ll_setattr(struct dentry *de, struct iattr *attr)
2119 int mode = de->d_inode->i_mode;
2120 enum op_xvalid xvalid = 0;
2123 rc = llcrypt_prepare_setattr(de, attr);
2127 if ((attr->ia_valid & (ATTR_CTIME|ATTR_SIZE|ATTR_MODE)) ==
2128 (ATTR_CTIME|ATTR_SIZE|ATTR_MODE))
2129 xvalid |= OP_XVALID_OWNEROVERRIDE;
2131 if (((attr->ia_valid & (ATTR_MODE|ATTR_FORCE|ATTR_SIZE)) ==
2132 (ATTR_SIZE|ATTR_MODE)) &&
2133 (((mode & S_ISUID) && !(attr->ia_mode & S_ISUID)) ||
2134 (((mode & (S_ISGID|S_IXGRP)) == (S_ISGID|S_IXGRP)) &&
2135 !(attr->ia_mode & S_ISGID))))
2136 attr->ia_valid |= ATTR_FORCE;
2138 if ((attr->ia_valid & ATTR_MODE) &&
2140 !(attr->ia_mode & S_ISUID) &&
2141 !(attr->ia_valid & ATTR_KILL_SUID))
2142 attr->ia_valid |= ATTR_KILL_SUID;
2144 if ((attr->ia_valid & ATTR_MODE) &&
2145 ((mode & (S_ISGID|S_IXGRP)) == (S_ISGID|S_IXGRP)) &&
2146 !(attr->ia_mode & S_ISGID) &&
2147 !(attr->ia_valid & ATTR_KILL_SGID))
2148 attr->ia_valid |= ATTR_KILL_SGID;
2150 return ll_setattr_raw(de, attr, xvalid, false);
2153 int ll_statfs_internal(struct ll_sb_info *sbi, struct obd_statfs *osfs,
2156 struct obd_statfs obd_osfs = { 0 };
2161 max_age = ktime_get_seconds() - sbi->ll_statfs_max_age;
2163 if (sbi->ll_flags & LL_SBI_LAZYSTATFS)
2164 flags |= OBD_STATFS_NODELAY;
2166 rc = obd_statfs(NULL, sbi->ll_md_exp, osfs, max_age, flags);
2170 osfs->os_type = LL_SUPER_MAGIC;
2172 CDEBUG(D_SUPER, "MDC blocks %llu/%llu objects %llu/%llu\n",
2173 osfs->os_bavail, osfs->os_blocks, osfs->os_ffree, osfs->os_files);
2175 if (osfs->os_state & OS_STATFS_SUM)
2178 rc = obd_statfs(NULL, sbi->ll_dt_exp, &obd_osfs, max_age, flags);
2179 if (rc) /* Possibly a filesystem with no OSTs. Report MDT totals. */
2182 CDEBUG(D_SUPER, "OSC blocks %llu/%llu objects %llu/%llu\n",
2183 obd_osfs.os_bavail, obd_osfs.os_blocks, obd_osfs.os_ffree,
2186 osfs->os_bsize = obd_osfs.os_bsize;
2187 osfs->os_blocks = obd_osfs.os_blocks;
2188 osfs->os_bfree = obd_osfs.os_bfree;
2189 osfs->os_bavail = obd_osfs.os_bavail;
2191 /* If we have _some_ OSTs, but don't have as many free objects on the
2192 * OSTs as inodes on the MDTs, reduce the reported number of inodes
2193 * to compensate, so that the "inodes in use" number is correct.
2194 * This should be kept in sync with lod_statfs() behaviour.
2196 if (obd_osfs.os_files && obd_osfs.os_ffree < osfs->os_ffree) {
2197 osfs->os_files = (osfs->os_files - osfs->os_ffree) +
2199 osfs->os_ffree = obd_osfs.os_ffree;
2206 int ll_statfs(struct dentry *de, struct kstatfs *sfs)
2208 struct super_block *sb = de->d_sb;
2209 struct obd_statfs osfs;
2210 __u64 fsid = huge_encode_dev(sb->s_dev);
2211 ktime_t kstart = ktime_get();
2214 CDEBUG(D_VFSTRACE, "VFS Op:sb=%s (%p)\n", sb->s_id, sb);
2216 /* Some amount of caching on the client is allowed */
2217 rc = ll_statfs_internal(ll_s2sbi(sb), &osfs, OBD_STATFS_SUM);
2221 statfs_unpack(sfs, &osfs);
2223 /* We need to downshift for all 32-bit kernels, because we can't
2224 * tell if the kernel is being called via sys_statfs64() or not.
2225 * Stop before overflowing f_bsize - in which case it is better
2226 * to just risk EOVERFLOW if caller is using old sys_statfs(). */
2227 if (sizeof(long) < 8) {
2228 while (osfs.os_blocks > ~0UL && sfs->f_bsize < 0x40000000) {
2231 osfs.os_blocks >>= 1;
2232 osfs.os_bfree >>= 1;
2233 osfs.os_bavail >>= 1;
2237 sfs->f_blocks = osfs.os_blocks;
2238 sfs->f_bfree = osfs.os_bfree;
2239 sfs->f_bavail = osfs.os_bavail;
2240 sfs->f_fsid.val[0] = (__u32)fsid;
2241 sfs->f_fsid.val[1] = (__u32)(fsid >> 32);
2243 ll_stats_ops_tally(ll_s2sbi(sb), LPROC_LL_STATFS,
2244 ktime_us_delta(ktime_get(), kstart));
2249 void ll_inode_size_lock(struct inode *inode)
2251 struct ll_inode_info *lli;
2253 LASSERT(!S_ISDIR(inode->i_mode));
2255 lli = ll_i2info(inode);
2256 mutex_lock(&lli->lli_size_mutex);
2259 void ll_inode_size_unlock(struct inode *inode)
2261 struct ll_inode_info *lli;
2263 lli = ll_i2info(inode);
2264 mutex_unlock(&lli->lli_size_mutex);
2267 void ll_update_inode_flags(struct inode *inode, int ext_flags)
2269 /* do not clear encryption flag */
2270 ext_flags |= ll_inode_to_ext_flags(inode->i_flags) & LUSTRE_ENCRYPT_FL;
2271 inode->i_flags = ll_ext_to_inode_flags(ext_flags);
2272 if (ext_flags & LUSTRE_PROJINHERIT_FL)
2273 ll_file_set_flag(ll_i2info(inode), LLIF_PROJECT_INHERIT);
2275 ll_file_clear_flag(ll_i2info(inode), LLIF_PROJECT_INHERIT);
2278 int ll_update_inode(struct inode *inode, struct lustre_md *md)
2280 struct ll_inode_info *lli = ll_i2info(inode);
2281 struct mdt_body *body = md->body;
2282 struct ll_sb_info *sbi = ll_i2sbi(inode);
2285 if (body->mbo_valid & OBD_MD_FLEASIZE) {
2286 rc = cl_file_inode_init(inode, md);
2291 if (S_ISDIR(inode->i_mode)) {
2292 rc = ll_update_lsm_md(inode, md);
2297 if (body->mbo_valid & OBD_MD_FLACL)
2298 lli_replace_acl(lli, md);
2300 inode->i_ino = cl_fid_build_ino(&body->mbo_fid1,
2301 sbi->ll_flags & LL_SBI_32BIT_API);
2302 inode->i_generation = cl_fid_build_gen(&body->mbo_fid1);
2304 if (body->mbo_valid & OBD_MD_FLATIME) {
2305 if (body->mbo_atime > inode->i_atime.tv_sec)
2306 inode->i_atime.tv_sec = body->mbo_atime;
2307 lli->lli_atime = body->mbo_atime;
2310 if (body->mbo_valid & OBD_MD_FLMTIME) {
2311 if (body->mbo_mtime > inode->i_mtime.tv_sec) {
2313 "setting ino %lu mtime from %lld to %llu\n",
2314 inode->i_ino, (s64)inode->i_mtime.tv_sec,
2316 inode->i_mtime.tv_sec = body->mbo_mtime;
2318 lli->lli_mtime = body->mbo_mtime;
2321 if (body->mbo_valid & OBD_MD_FLCTIME) {
2322 if (body->mbo_ctime > inode->i_ctime.tv_sec)
2323 inode->i_ctime.tv_sec = body->mbo_ctime;
2324 lli->lli_ctime = body->mbo_ctime;
2327 if (body->mbo_valid & OBD_MD_FLBTIME)
2328 lli->lli_btime = body->mbo_btime;
2330 /* Clear i_flags to remove S_NOSEC before permissions are updated */
2331 if (body->mbo_valid & OBD_MD_FLFLAGS)
2332 ll_update_inode_flags(inode, body->mbo_flags);
2333 if (body->mbo_valid & OBD_MD_FLMODE)
2334 inode->i_mode = (inode->i_mode & S_IFMT) |
2335 (body->mbo_mode & ~S_IFMT);
2337 if (body->mbo_valid & OBD_MD_FLTYPE)
2338 inode->i_mode = (inode->i_mode & ~S_IFMT) |
2339 (body->mbo_mode & S_IFMT);
2341 LASSERT(inode->i_mode != 0);
2342 if (body->mbo_valid & OBD_MD_FLUID)
2343 inode->i_uid = make_kuid(&init_user_ns, body->mbo_uid);
2344 if (body->mbo_valid & OBD_MD_FLGID)
2345 inode->i_gid = make_kgid(&init_user_ns, body->mbo_gid);
2346 if (body->mbo_valid & OBD_MD_FLPROJID)
2347 lli->lli_projid = body->mbo_projid;
2348 if (body->mbo_valid & OBD_MD_FLNLINK)
2349 set_nlink(inode, body->mbo_nlink);
2350 if (body->mbo_valid & OBD_MD_FLRDEV)
2351 inode->i_rdev = old_decode_dev(body->mbo_rdev);
2353 if (body->mbo_valid & OBD_MD_FLID) {
2354 /* FID shouldn't be changed! */
2355 if (fid_is_sane(&lli->lli_fid)) {
2356 LASSERTF(lu_fid_eq(&lli->lli_fid, &body->mbo_fid1),
2357 "Trying to change FID "DFID
2358 " to the "DFID", inode "DFID"(%p)\n",
2359 PFID(&lli->lli_fid), PFID(&body->mbo_fid1),
2360 PFID(ll_inode2fid(inode)), inode);
2362 lli->lli_fid = body->mbo_fid1;
2366 LASSERT(fid_seq(&lli->lli_fid) != 0);
2368 lli->lli_attr_valid = body->mbo_valid;
2369 if (body->mbo_valid & OBD_MD_FLSIZE) {
2370 i_size_write(inode, body->mbo_size);
2372 CDEBUG(D_VFSTRACE, "inode="DFID", updating i_size %llu\n",
2373 PFID(ll_inode2fid(inode)),
2374 (unsigned long long)body->mbo_size);
2376 if (body->mbo_valid & OBD_MD_FLBLOCKS)
2377 inode->i_blocks = body->mbo_blocks;
2379 if (body->mbo_valid & OBD_MD_FLLAZYSIZE)
2380 lli->lli_lazysize = body->mbo_size;
2381 if (body->mbo_valid & OBD_MD_FLLAZYBLOCKS)
2382 lli->lli_lazyblocks = body->mbo_blocks;
2385 if (body->mbo_valid & OBD_MD_TSTATE) {
2386 /* Set LLIF_FILE_RESTORING if restore ongoing and
2387 * clear it when done to ensure to start again
2388 * glimpsing updated attrs
2390 if (body->mbo_t_state & MS_RESTORE)
2391 ll_file_set_flag(lli, LLIF_FILE_RESTORING);
2393 ll_file_clear_flag(lli, LLIF_FILE_RESTORING);
2399 int ll_read_inode2(struct inode *inode, void *opaque)
2401 struct lustre_md *md = opaque;
2402 struct ll_inode_info *lli = ll_i2info(inode);
2406 CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p)\n",
2407 PFID(&lli->lli_fid), inode);
2409 /* Core attributes from the MDS first. This is a new inode, and
2410 * the VFS doesn't zero times in the core inode so we have to do
2411 * it ourselves. They will be overwritten by either MDS or OST
2412 * attributes - we just need to make sure they aren't newer.
2414 inode->i_mtime.tv_sec = 0;
2415 inode->i_atime.tv_sec = 0;
2416 inode->i_ctime.tv_sec = 0;
2418 rc = ll_update_inode(inode, md);
2422 /* OIDEBUG(inode); */
2424 #ifdef HAVE_BACKING_DEV_INFO
2425 /* initializing backing dev info. */
2426 inode->i_mapping->backing_dev_info = &s2lsi(inode->i_sb)->lsi_bdi;
2428 if (S_ISREG(inode->i_mode)) {
2429 struct ll_sb_info *sbi = ll_i2sbi(inode);
2430 inode->i_op = &ll_file_inode_operations;
2431 inode->i_fop = sbi->ll_fop;
2432 inode->i_mapping->a_ops = (struct address_space_operations *)&ll_aops;
2434 } else if (S_ISDIR(inode->i_mode)) {
2435 inode->i_op = &ll_dir_inode_operations;
2436 inode->i_fop = &ll_dir_operations;
2438 } else if (S_ISLNK(inode->i_mode)) {
2439 inode->i_op = &ll_fast_symlink_inode_operations;
2442 inode->i_op = &ll_special_inode_operations;
2444 init_special_inode(inode, inode->i_mode,
2453 void ll_delete_inode(struct inode *inode)
2455 struct ll_inode_info *lli = ll_i2info(inode);
2456 struct address_space *mapping = &inode->i_data;
2457 unsigned long nrpages;
2458 unsigned long flags;
2462 if (S_ISREG(inode->i_mode) && lli->lli_clob != NULL) {
2463 /* It is last chance to write out dirty pages,
2464 * otherwise we may lose data while umount.
2466 * If i_nlink is 0 then just discard data. This is safe because
2467 * local inode gets i_nlink 0 from server only for the last
2468 * unlink, so that file is not opened somewhere else
2470 cl_sync_file_range(inode, 0, OBD_OBJECT_EOF, inode->i_nlink ?
2471 CL_FSYNC_LOCAL : CL_FSYNC_DISCARD, 1);
2473 truncate_inode_pages_final(mapping);
2475 /* Workaround for LU-118: Note nrpages may not be totally updated when
2476 * truncate_inode_pages() returns, as there can be a page in the process
2477 * of deletion (inside __delete_from_page_cache()) in the specified
2478 * range. Thus mapping->nrpages can be non-zero when this function
2479 * returns even after truncation of the whole mapping. Only do this if
2480 * npages isn't already zero.
2482 nrpages = mapping->nrpages;
2484 ll_xa_lock_irqsave(&mapping->i_pages, flags);
2485 nrpages = mapping->nrpages;
2486 ll_xa_unlock_irqrestore(&mapping->i_pages, flags);
2487 } /* Workaround end */
2489 LASSERTF(nrpages == 0, "%s: inode="DFID"(%p) nrpages=%lu, "
2490 "see https://jira.whamcloud.com/browse/LU-118\n",
2491 ll_i2sbi(inode)->ll_fsname,
2492 PFID(ll_inode2fid(inode)), inode, nrpages);
2494 ll_clear_inode(inode);
2500 int ll_iocontrol(struct inode *inode, struct file *file,
2501 unsigned int cmd, unsigned long arg)
2503 struct ll_sb_info *sbi = ll_i2sbi(inode);
2504 struct ptlrpc_request *req = NULL;
2509 case FS_IOC_GETFLAGS: {
2510 struct mdt_body *body;
2511 struct md_op_data *op_data;
2513 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL,
2514 0, 0, LUSTRE_OPC_ANY,
2516 if (IS_ERR(op_data))
2517 RETURN(PTR_ERR(op_data));
2519 op_data->op_valid = OBD_MD_FLFLAGS;
2520 rc = md_getattr(sbi->ll_md_exp, op_data, &req);
2521 ll_finish_md_op_data(op_data);
2523 CERROR("%s: failure inode "DFID": rc = %d\n",
2524 sbi->ll_md_exp->exp_obd->obd_name,
2525 PFID(ll_inode2fid(inode)), rc);
2529 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
2531 flags = body->mbo_flags;
2533 ptlrpc_req_finished(req);
2535 RETURN(put_user(flags, (int __user *)arg));
2537 case FS_IOC_SETFLAGS: {
2539 struct md_op_data *op_data;
2540 struct cl_object *obj;
2541 struct fsxattr fa = { 0 };
2543 if (get_user(flags, (int __user *)arg))
2546 fa.fsx_projid = ll_i2info(inode)->lli_projid;
2547 if (flags & LUSTRE_PROJINHERIT_FL)
2548 fa.fsx_xflags = FS_XFLAG_PROJINHERIT;
2550 rc = ll_ioctl_check_project(inode, &fa);
2554 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
2555 LUSTRE_OPC_ANY, NULL);
2556 if (IS_ERR(op_data))
2557 RETURN(PTR_ERR(op_data));
2559 op_data->op_attr_flags = flags;
2560 op_data->op_xvalid |= OP_XVALID_FLAGS;
2561 rc = md_setattr(sbi->ll_md_exp, op_data, NULL, 0, &req);
2562 ll_finish_md_op_data(op_data);
2563 ptlrpc_req_finished(req);
2567 ll_update_inode_flags(inode, flags);
2569 obj = ll_i2info(inode)->lli_clob;
2573 OBD_ALLOC_PTR(attr);
2577 rc = cl_setattr_ost(obj, attr, OP_XVALID_FLAGS, flags);
2589 int ll_flush_ctx(struct inode *inode)
2591 struct ll_sb_info *sbi = ll_i2sbi(inode);
2593 CDEBUG(D_SEC, "flush context for user %d\n",
2594 from_kuid(&init_user_ns, current_uid()));
2596 obd_set_info_async(NULL, sbi->ll_md_exp,
2597 sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
2599 obd_set_info_async(NULL, sbi->ll_dt_exp,
2600 sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
2605 /* umount -f client means force down, don't save state */
2606 void ll_umount_begin(struct super_block *sb)
2608 struct ll_sb_info *sbi = ll_s2sbi(sb);
2609 struct obd_device *obd;
2610 struct obd_ioctl_data *ioc_data;
2614 CDEBUG(D_VFSTRACE, "VFS Op: superblock %p count %d active %d\n", sb,
2615 sb->s_count, atomic_read(&sb->s_active));
2617 obd = class_exp2obd(sbi->ll_md_exp);
2619 CERROR("Invalid MDC connection handle %#llx\n",
2620 sbi->ll_md_exp->exp_handle.h_cookie);
2626 obd = class_exp2obd(sbi->ll_dt_exp);
2628 CERROR("Invalid LOV connection handle %#llx\n",
2629 sbi->ll_dt_exp->exp_handle.h_cookie);
2635 OBD_ALLOC_PTR(ioc_data);
2637 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_md_exp,
2638 sizeof *ioc_data, ioc_data, NULL);
2640 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_dt_exp,
2641 sizeof *ioc_data, ioc_data, NULL);
2643 OBD_FREE_PTR(ioc_data);
2646 /* Really, we'd like to wait until there are no requests outstanding,
2647 * and then continue. For now, we just periodically checking for vfs
2648 * to decrement mnt_cnt and hope to finish it within 10sec.
2652 !may_umount(sbi->ll_mnt.mnt)) {
2660 int ll_remount_fs(struct super_block *sb, int *flags, char *data)
2662 struct ll_sb_info *sbi = ll_s2sbi(sb);
2663 char *profilenm = get_profile_name(sb);
2667 if ((*flags & MS_RDONLY) != (sb->s_flags & SB_RDONLY)) {
2668 read_only = *flags & MS_RDONLY;
2669 err = obd_set_info_async(NULL, sbi->ll_md_exp,
2670 sizeof(KEY_READ_ONLY),
2671 KEY_READ_ONLY, sizeof(read_only),
2674 LCONSOLE_WARN("Failed to remount %s %s (%d)\n",
2675 profilenm, read_only ?
2676 "read-only" : "read-write", err);
2681 sb->s_flags |= SB_RDONLY;
2683 sb->s_flags &= ~SB_RDONLY;
2685 if (sbi->ll_flags & LL_SBI_VERBOSE)
2686 LCONSOLE_WARN("Remounted %s %s\n", profilenm,
2687 read_only ? "read-only" : "read-write");
2693 * Cleanup the open handle that is cached on MDT-side.
2695 * For open case, the client side open handling thread may hit error
2696 * after the MDT grant the open. Under such case, the client should
2697 * send close RPC to the MDT as cleanup; otherwise, the open handle
2698 * on the MDT will be leaked there until the client umount or evicted.
2700 * In further, if someone unlinked the file, because the open handle
2701 * holds the reference on such file/object, then it will block the
2702 * subsequent threads that want to locate such object via FID.
2704 * \param[in] sb super block for this file-system
2705 * \param[in] open_req pointer to the original open request
2707 void ll_open_cleanup(struct super_block *sb, struct ptlrpc_request *open_req)
2709 struct mdt_body *body;
2710 struct md_op_data *op_data;
2711 struct ptlrpc_request *close_req = NULL;
2712 struct obd_export *exp = ll_s2sbi(sb)->ll_md_exp;
2715 body = req_capsule_server_get(&open_req->rq_pill, &RMF_MDT_BODY);
2716 OBD_ALLOC_PTR(op_data);
2717 if (op_data == NULL) {
2718 CWARN("%s: cannot allocate op_data to release open handle for "
2719 DFID"\n", ll_s2sbi(sb)->ll_fsname, PFID(&body->mbo_fid1));
2724 op_data->op_fid1 = body->mbo_fid1;
2725 op_data->op_open_handle = body->mbo_open_handle;
2726 op_data->op_mod_time = ktime_get_real_seconds();
2727 md_close(exp, op_data, NULL, &close_req);
2728 ptlrpc_req_finished(close_req);
2729 ll_finish_md_op_data(op_data);
2734 int ll_prep_inode(struct inode **inode, struct ptlrpc_request *req,
2735 struct super_block *sb, struct lookup_intent *it)
2737 struct ll_sb_info *sbi = NULL;
2738 struct lustre_md md = { NULL };
2739 bool default_lmv_deleted = false;
2744 LASSERT(*inode || sb);
2745 sbi = sb ? ll_s2sbi(sb) : ll_i2sbi(*inode);
2746 rc = md_get_lustre_md(sbi->ll_md_exp, req, sbi->ll_dt_exp,
2747 sbi->ll_md_exp, &md);
2752 * clear default_lmv only if intent_getattr reply doesn't contain it.
2753 * but it needs to be done after iget, check this early because
2754 * ll_update_lsm_md() may change md.
2756 if (it && (it->it_op & (IT_LOOKUP | IT_GETATTR)) &&
2757 S_ISDIR(md.body->mbo_mode) && !md.default_lmv)
2758 default_lmv_deleted = true;
2761 rc = ll_update_inode(*inode, &md);
2765 LASSERT(sb != NULL);
2768 * At this point server returns to client's same fid as client
2769 * generated for creating. So using ->fid1 is okay here.
2771 if (!fid_is_sane(&md.body->mbo_fid1)) {
2772 CERROR("%s: Fid is insane "DFID"\n",
2774 PFID(&md.body->mbo_fid1));
2775 GOTO(out, rc = -EINVAL);
2778 *inode = ll_iget(sb, cl_fid_build_ino(&md.body->mbo_fid1,
2779 sbi->ll_flags & LL_SBI_32BIT_API),
2781 if (IS_ERR(*inode)) {
2783 rc = IS_ERR(*inode) ? PTR_ERR(*inode) : -ENOMEM;
2785 CERROR("new_inode -fatal: rc %d\n", rc);
2790 /* Handling piggyback layout lock.
2791 * Layout lock can be piggybacked by getattr and open request.
2792 * The lsm can be applied to inode only if it comes with a layout lock
2793 * otherwise correct layout may be overwritten, for example:
2794 * 1. proc1: mdt returns a lsm but not granting layout
2795 * 2. layout was changed by another client
2796 * 3. proc2: refresh layout and layout lock granted
2797 * 4. proc1: to apply a stale layout */
2798 if (it != NULL && it->it_lock_mode != 0) {
2799 struct lustre_handle lockh;
2800 struct ldlm_lock *lock;
2802 lockh.cookie = it->it_lock_handle;
2803 lock = ldlm_handle2lock(&lockh);
2804 LASSERT(lock != NULL);
2805 if (ldlm_has_layout(lock)) {
2806 struct cl_object_conf conf;
2808 memset(&conf, 0, sizeof(conf));
2809 conf.coc_opc = OBJECT_CONF_SET;
2810 conf.coc_inode = *inode;
2811 conf.coc_lock = lock;
2812 conf.u.coc_layout = md.layout;
2813 (void)ll_layout_conf(*inode, &conf);
2815 LDLM_LOCK_PUT(lock);
2818 if (default_lmv_deleted)
2819 ll_update_default_lsm_md(*inode, &md);
2824 /* cleanup will be done if necessary */
2825 md_free_lustre_md(sbi->ll_md_exp, &md);
2827 if (rc != 0 && it != NULL && it->it_op & IT_OPEN) {
2828 ll_intent_drop_lock(it);
2829 ll_open_cleanup(sb != NULL ? sb : (*inode)->i_sb, req);
2835 int ll_obd_statfs(struct inode *inode, void __user *arg)
2837 struct ll_sb_info *sbi = NULL;
2838 struct obd_export *exp;
2840 struct obd_ioctl_data *data = NULL;
2844 if (!inode || !(sbi = ll_i2sbi(inode)))
2845 GOTO(out_statfs, rc = -EINVAL);
2847 rc = obd_ioctl_getdata(&buf, &len, arg);
2849 GOTO(out_statfs, rc);
2852 if (!data->ioc_inlbuf1 || !data->ioc_inlbuf2 ||
2853 !data->ioc_pbuf1 || !data->ioc_pbuf2)
2854 GOTO(out_statfs, rc = -EINVAL);
2856 if (data->ioc_inllen1 != sizeof(__u32) ||
2857 data->ioc_inllen2 != sizeof(__u32) ||
2858 data->ioc_plen1 != sizeof(struct obd_statfs) ||
2859 data->ioc_plen2 != sizeof(struct obd_uuid))
2860 GOTO(out_statfs, rc = -EINVAL);
2862 memcpy(&type, data->ioc_inlbuf1, sizeof(__u32));
2863 if (type & LL_STATFS_LMV)
2864 exp = sbi->ll_md_exp;
2865 else if (type & LL_STATFS_LOV)
2866 exp = sbi->ll_dt_exp;
2868 GOTO(out_statfs, rc = -ENODEV);
2870 rc = obd_iocontrol(IOC_OBD_STATFS, exp, len, buf, NULL);
2872 GOTO(out_statfs, rc);
2874 OBD_FREE_LARGE(buf, len);
2879 * this is normally called in ll_fini_md_op_data(), but sometimes it needs to
2880 * be called early to avoid deadlock.
2882 void ll_unlock_md_op_lsm(struct md_op_data *op_data)
2884 if (op_data->op_mea2_sem) {
2885 up_read_non_owner(op_data->op_mea2_sem);
2886 op_data->op_mea2_sem = NULL;
2889 if (op_data->op_mea1_sem) {
2890 up_read_non_owner(op_data->op_mea1_sem);
2891 op_data->op_mea1_sem = NULL;
2895 /* this function prepares md_op_data hint for passing it down to MD stack. */
2896 struct md_op_data *ll_prep_md_op_data(struct md_op_data *op_data,
2897 struct inode *i1, struct inode *i2,
2898 const char *name, size_t namelen,
2899 __u32 mode, enum md_op_code opc,
2902 LASSERT(i1 != NULL);
2905 /* Do not reuse namelen for something else. */
2907 return ERR_PTR(-EINVAL);
2909 if (namelen > ll_i2sbi(i1)->ll_namelen)
2910 return ERR_PTR(-ENAMETOOLONG);
2912 if (!lu_name_is_valid_2(name, namelen))
2913 return ERR_PTR(-EINVAL);
2916 if (op_data == NULL)
2917 OBD_ALLOC_PTR(op_data);
2919 if (op_data == NULL)
2920 return ERR_PTR(-ENOMEM);
2922 ll_i2gids(op_data->op_suppgids, i1, i2);
2923 op_data->op_fid1 = *ll_inode2fid(i1);
2924 op_data->op_code = opc;
2926 if (S_ISDIR(i1->i_mode)) {
2927 down_read_non_owner(&ll_i2info(i1)->lli_lsm_sem);
2928 op_data->op_mea1_sem = &ll_i2info(i1)->lli_lsm_sem;
2929 op_data->op_mea1 = ll_i2info(i1)->lli_lsm_md;
2930 op_data->op_default_mea1 = ll_i2info(i1)->lli_default_lsm_md;
2934 op_data->op_fid2 = *ll_inode2fid(i2);
2935 if (S_ISDIR(i2->i_mode)) {
2937 /* i2 is typically a child of i1, and MUST be
2938 * further from the root to avoid deadlocks.
2940 down_read_non_owner(&ll_i2info(i2)->lli_lsm_sem);
2941 op_data->op_mea2_sem =
2942 &ll_i2info(i2)->lli_lsm_sem;
2944 op_data->op_mea2 = ll_i2info(i2)->lli_lsm_md;
2947 fid_zero(&op_data->op_fid2);
2950 if (ll_i2sbi(i1)->ll_flags & LL_SBI_64BIT_HASH)
2951 op_data->op_cli_flags |= CLI_HASH64;
2953 if (ll_need_32bit_api(ll_i2sbi(i1)))
2954 op_data->op_cli_flags |= CLI_API32;
2956 op_data->op_name = name;
2957 op_data->op_namelen = namelen;
2958 op_data->op_mode = mode;
2959 op_data->op_mod_time = ktime_get_real_seconds();
2960 op_data->op_fsuid = from_kuid(&init_user_ns, current_fsuid());
2961 op_data->op_fsgid = from_kgid(&init_user_ns, current_fsgid());
2962 op_data->op_cap = cfs_curproc_cap_pack();
2963 op_data->op_mds = 0;
2964 if ((opc == LUSTRE_OPC_CREATE) && (name != NULL) &&
2965 filename_is_volatile(name, namelen, &op_data->op_mds)) {
2966 op_data->op_bias |= MDS_CREATE_VOLATILE;
2968 op_data->op_data = data;
2973 void ll_finish_md_op_data(struct md_op_data *op_data)
2975 ll_unlock_md_op_lsm(op_data);
2976 security_release_secctx(op_data->op_file_secctx,
2977 op_data->op_file_secctx_size);
2978 llcrypt_free_ctx(op_data->op_file_encctx, op_data->op_file_encctx_size);
2979 OBD_FREE_PTR(op_data);
2982 int ll_show_options(struct seq_file *seq, struct dentry *dentry)
2984 struct ll_sb_info *sbi;
2986 LASSERT(seq && dentry);
2987 sbi = ll_s2sbi(dentry->d_sb);
2989 if (sbi->ll_flags & LL_SBI_NOLCK)
2990 seq_puts(seq, ",nolock");
2992 /* "flock" is the default since 2.13, but it wasn't for many years,
2993 * so it is still useful to print this to show it is enabled.
2994 * Start to print "noflock" so it is now clear when flock is disabled.
2996 if (sbi->ll_flags & LL_SBI_FLOCK)
2997 seq_puts(seq, ",flock");
2998 else if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
2999 seq_puts(seq, ",localflock");
3001 seq_puts(seq, ",noflock");
3003 if (sbi->ll_flags & LL_SBI_USER_XATTR)
3004 seq_puts(seq, ",user_xattr");
3006 if (sbi->ll_flags & LL_SBI_LAZYSTATFS)
3007 seq_puts(seq, ",lazystatfs");
3009 if (sbi->ll_flags & LL_SBI_USER_FID2PATH)
3010 seq_puts(seq, ",user_fid2path");
3012 if (sbi->ll_flags & LL_SBI_ALWAYS_PING)
3013 seq_puts(seq, ",always_ping");
3015 if (ll_sbi_has_test_dummy_encryption(sbi))
3016 seq_puts(seq, ",test_dummy_encryption");
3018 if (ll_sbi_has_encrypt(sbi))
3019 seq_puts(seq, ",encrypt");
3021 seq_puts(seq, ",noencrypt");
3027 * Get obd name by cmd, and copy out to user space
3029 int ll_get_obd_name(struct inode *inode, unsigned int cmd, unsigned long arg)
3031 struct ll_sb_info *sbi = ll_i2sbi(inode);
3032 struct obd_device *obd;
3035 if (cmd == OBD_IOC_GETDTNAME)
3036 obd = class_exp2obd(sbi->ll_dt_exp);
3037 else if (cmd == OBD_IOC_GETMDNAME)
3038 obd = class_exp2obd(sbi->ll_md_exp);
3045 if (copy_to_user((void __user *)arg, obd->obd_name,
3046 strlen(obd->obd_name) + 1))
3052 static char* ll_d_path(struct dentry *dentry, char *buf, int bufsize)
3059 p.mnt = current->fs->root.mnt;
3061 path = d_path(&p, buf, bufsize);
3066 void ll_dirty_page_discard_warn(struct page *page, int ioret)
3068 char *buf, *path = NULL;
3069 struct dentry *dentry = NULL;
3070 struct inode *inode = page->mapping->host;
3072 /* this can be called inside spin lock so use GFP_ATOMIC. */
3073 buf = (char *)__get_free_page(GFP_ATOMIC);
3075 dentry = d_find_alias(page->mapping->host);
3077 path = ll_d_path(dentry, buf, PAGE_SIZE);
3080 /* The below message is checked in recovery-small.sh test_24b */
3082 "%s: dirty page discard: %s/fid: "DFID"/%s may get corrupted "
3083 "(rc %d)\n", ll_i2sbi(inode)->ll_fsname,
3084 s2lsi(page->mapping->host->i_sb)->lsi_lmd->lmd_dev,
3085 PFID(ll_inode2fid(inode)),
3086 (path && !IS_ERR(path)) ? path : "", ioret);
3092 free_page((unsigned long)buf);
3095 ssize_t ll_copy_user_md(const struct lov_user_md __user *md,
3096 struct lov_user_md **kbuf)
3098 struct lov_user_md lum;
3102 if (copy_from_user(&lum, md, sizeof(lum)))
3105 lum_size = ll_lov_user_md_size(&lum);
3109 OBD_ALLOC_LARGE(*kbuf, lum_size);
3113 if (copy_from_user(*kbuf, md, lum_size) != 0) {
3114 OBD_FREE_LARGE(*kbuf, lum_size);
3122 * Compute llite root squash state after a change of root squash
3123 * configuration setting or add/remove of a lnet nid
3125 void ll_compute_rootsquash_state(struct ll_sb_info *sbi)
3127 struct root_squash_info *squash = &sbi->ll_squash;
3130 struct lnet_process_id id;
3132 /* Update norootsquash flag */
3133 spin_lock(&squash->rsi_lock);
3134 if (list_empty(&squash->rsi_nosquash_nids))
3135 sbi->ll_flags &= ~LL_SBI_NOROOTSQUASH;
3137 /* Do not apply root squash as soon as one of our NIDs is
3138 * in the nosquash_nids list */
3141 while (LNetGetId(i++, &id) != -ENOENT) {
3142 if (id.nid == LNET_NID_LO_0)
3144 if (cfs_match_nid(id.nid, &squash->rsi_nosquash_nids)) {
3150 sbi->ll_flags |= LL_SBI_NOROOTSQUASH;
3152 sbi->ll_flags &= ~LL_SBI_NOROOTSQUASH;
3154 spin_unlock(&squash->rsi_lock);
3158 * Parse linkea content to extract information about a given hardlink
3160 * \param[in] ldata - Initialized linkea data
3161 * \param[in] linkno - Link identifier
3162 * \param[out] parent_fid - The entry's parent FID
3163 * \param[out] ln - Entry name destination buffer
3165 * \retval 0 on success
3166 * \retval Appropriate negative error code on failure
3168 static int ll_linkea_decode(struct linkea_data *ldata, unsigned int linkno,
3169 struct lu_fid *parent_fid, struct lu_name *ln)
3175 rc = linkea_init_with_rec(ldata);
3179 if (linkno >= ldata->ld_leh->leh_reccount)
3180 /* beyond last link */
3183 linkea_first_entry(ldata);
3184 for (idx = 0; ldata->ld_lee != NULL; idx++) {
3185 linkea_entry_unpack(ldata->ld_lee, &ldata->ld_reclen, ln,
3190 linkea_next_entry(ldata);
3200 * Get parent FID and name of an identified link. Operation is performed for
3201 * a given link number, letting the caller iterate over linkno to list one or
3202 * all links of an entry.
3204 * \param[in] file - File descriptor against which to perform the operation
3205 * \param[in,out] arg - User-filled structure containing the linkno to operate
3206 * on and the available size. It is eventually filled with
3207 * the requested information or left untouched on error
3209 * \retval - 0 on success
3210 * \retval - Appropriate negative error code on failure
3212 int ll_getparent(struct file *file, struct getparent __user *arg)
3214 struct inode *inode = file_inode(file);
3215 struct linkea_data *ldata;
3216 struct lu_buf buf = LU_BUF_NULL;
3218 struct lu_fid parent_fid;
3225 if (!cfs_capable(CFS_CAP_DAC_READ_SEARCH) &&
3226 !(ll_i2sbi(inode)->ll_flags & LL_SBI_USER_FID2PATH))
3229 if (get_user(name_size, &arg->gp_name_size))
3232 if (get_user(linkno, &arg->gp_linkno))
3235 if (name_size > PATH_MAX)
3238 OBD_ALLOC(ldata, sizeof(*ldata));
3242 rc = linkea_data_new(ldata, &buf);
3244 GOTO(ldata_free, rc);
3246 rc = ll_xattr_list(inode, XATTR_NAME_LINK, XATTR_TRUSTED_T, buf.lb_buf,
3247 buf.lb_len, OBD_MD_FLXATTR);
3251 rc = ll_linkea_decode(ldata, linkno, &parent_fid, &ln);
3255 if (ln.ln_namelen >= name_size)
3256 GOTO(lb_free, rc = -EOVERFLOW);
3258 if (copy_to_user(&arg->gp_fid, &parent_fid, sizeof(arg->gp_fid)))
3259 GOTO(lb_free, rc = -EFAULT);
3261 if (copy_to_user(&arg->gp_name, ln.ln_name, ln.ln_namelen))
3262 GOTO(lb_free, rc = -EFAULT);
3264 if (put_user('\0', arg->gp_name + ln.ln_namelen))
3265 GOTO(lb_free, rc = -EFAULT);
3270 OBD_FREE(ldata, sizeof(*ldata));