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/
31 * lustre/llite/llite_lib.c
33 * Lustre Light Super operations
36 #define DEBUG_SUBSYSTEM S_LLITE
38 #include <linux/cpu.h>
39 #include <linux/module.h>
40 #include <linux/random.h>
41 #include <linux/statfs.h>
42 #include <linux/time.h>
43 #include <linux/file.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/fs_struct.h>
53 #ifndef HAVE_CPUS_READ_LOCK
54 #include <libcfs/linux/linux-cpu.h>
56 #include <libcfs/linux/linux-misc.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;
101 RETURN(ERR_PTR(-ENOMEM));
103 rc = pcc_super_init(&sbi->ll_pcc_super);
107 spin_lock_init(&sbi->ll_lock);
108 mutex_init(&sbi->ll_lco.lco_lock);
109 spin_lock_init(&sbi->ll_pp_extent_lock);
110 spin_lock_init(&sbi->ll_process_lock);
111 sbi->ll_rw_stats_on = 0;
112 sbi->ll_statfs_max_age = OBD_STATFS_CACHE_SECONDS;
115 pages = si.totalram - si.totalhigh;
116 lru_page_max = pages / 2;
118 sbi->ll_ra_info.ra_async_max_active = ll_get_ra_async_max_active();
119 sbi->ll_ra_info.ll_readahead_wq =
120 cfs_cpt_bind_workqueue("ll-readahead-wq", cfs_cpt_tab,
122 sbi->ll_ra_info.ra_async_max_active);
123 if (IS_ERR(sbi->ll_ra_info.ll_readahead_wq))
124 GOTO(out_pcc, rc = PTR_ERR(sbi->ll_ra_info.ll_readahead_wq));
126 /* initialize ll_cache data */
127 sbi->ll_cache = cl_cache_init(lru_page_max);
128 if (sbi->ll_cache == NULL)
129 GOTO(out_destroy_ra, rc = -ENOMEM);
131 /* initialize foreign symlink prefix path */
132 OBD_ALLOC(sbi->ll_foreign_symlink_prefix, sizeof("/mnt/"));
133 if (sbi->ll_foreign_symlink_prefix == NULL)
134 GOTO(out_destroy_ra, rc = -ENOMEM);
135 memcpy(sbi->ll_foreign_symlink_prefix, "/mnt/", sizeof("/mnt/"));
136 sbi->ll_foreign_symlink_prefix_size = sizeof("/mnt/");
138 /* initialize foreign symlink upcall path, none by default */
139 OBD_ALLOC(sbi->ll_foreign_symlink_upcall, sizeof("none"));
140 if (sbi->ll_foreign_symlink_upcall == NULL)
141 GOTO(out_destroy_ra, rc = -ENOMEM);
142 memcpy(sbi->ll_foreign_symlink_upcall, "none", sizeof("none"));
143 sbi->ll_foreign_symlink_upcall_items = NULL;
144 sbi->ll_foreign_symlink_upcall_nb_items = 0;
145 init_rwsem(&sbi->ll_foreign_symlink_sem);
146 /* foreign symlink support (LL_SBI_FOREIGN_SYMLINK in ll_flags)
147 * not enabled by default
150 sbi->ll_ra_info.ra_max_pages =
151 min(pages / 32, SBI_DEFAULT_READ_AHEAD_MAX);
152 sbi->ll_ra_info.ra_max_pages_per_file =
153 min(sbi->ll_ra_info.ra_max_pages / 4,
154 SBI_DEFAULT_READ_AHEAD_PER_FILE_MAX);
155 sbi->ll_ra_info.ra_async_pages_per_file_threshold =
156 sbi->ll_ra_info.ra_max_pages_per_file;
157 sbi->ll_ra_info.ra_range_pages = SBI_DEFAULT_RA_RANGE_PAGES;
158 sbi->ll_ra_info.ra_max_read_ahead_whole_pages = -1;
159 atomic_set(&sbi->ll_ra_info.ra_async_inflight, 0);
161 set_bit(LL_SBI_VERBOSE, sbi->ll_flags);
162 #ifdef ENABLE_CHECKSUM
163 set_bit(LL_SBI_CHECKSUM, sbi->ll_flags);
166 set_bit(LL_SBI_FLOCK, sbi->ll_flags);
169 #ifdef HAVE_LRU_RESIZE_SUPPORT
170 set_bit(LL_SBI_LRU_RESIZE, sbi->ll_flags);
172 set_bit(LL_SBI_LAZYSTATFS, sbi->ll_flags);
174 /* metadata statahead is enabled by default */
175 sbi->ll_sa_running_max = LL_SA_RUNNING_DEF;
176 sbi->ll_sa_max = LL_SA_RPC_DEF;
177 atomic_set(&sbi->ll_sa_total, 0);
178 atomic_set(&sbi->ll_sa_wrong, 0);
179 atomic_set(&sbi->ll_sa_running, 0);
180 atomic_set(&sbi->ll_agl_total, 0);
181 atomic_set(&sbi->ll_sa_hit_total, 0);
182 atomic_set(&sbi->ll_sa_miss_total, 0);
183 set_bit(LL_SBI_AGL_ENABLED, sbi->ll_flags);
184 set_bit(LL_SBI_FAST_READ, sbi->ll_flags);
185 set_bit(LL_SBI_TINY_WRITE, sbi->ll_flags);
186 set_bit(LL_SBI_PARALLEL_DIO, sbi->ll_flags);
187 ll_sbi_set_encrypt(sbi, true);
188 ll_sbi_set_name_encrypt(sbi, true);
191 sbi->ll_squash.rsi_uid = 0;
192 sbi->ll_squash.rsi_gid = 0;
193 INIT_LIST_HEAD(&sbi->ll_squash.rsi_nosquash_nids);
194 spin_lock_init(&sbi->ll_squash.rsi_lock);
196 /* Per-filesystem file heat */
197 sbi->ll_heat_decay_weight = SBI_DEFAULT_HEAT_DECAY_WEIGHT;
198 sbi->ll_heat_period_second = SBI_DEFAULT_HEAT_PERIOD_SECOND;
200 /* Per-fs open heat level before requesting open lock */
201 sbi->ll_oc_thrsh_count = SBI_DEFAULT_OPENCACHE_THRESHOLD_COUNT;
202 sbi->ll_oc_max_ms = SBI_DEFAULT_OPENCACHE_THRESHOLD_MAX_MS;
203 sbi->ll_oc_thrsh_ms = SBI_DEFAULT_OPENCACHE_THRESHOLD_MS;
206 if (sbi->ll_foreign_symlink_prefix)
207 OBD_FREE(sbi->ll_foreign_symlink_prefix, sizeof("/mnt/"));
209 cl_cache_decref(sbi->ll_cache);
210 sbi->ll_cache = NULL;
212 destroy_workqueue(sbi->ll_ra_info.ll_readahead_wq);
214 pcc_super_fini(&sbi->ll_pcc_super);
220 static void ll_free_sbi(struct super_block *sb)
222 struct ll_sb_info *sbi = ll_s2sbi(sb);
226 if (!list_empty(&sbi->ll_squash.rsi_nosquash_nids))
227 cfs_free_nidlist(&sbi->ll_squash.rsi_nosquash_nids);
228 if (sbi->ll_ra_info.ll_readahead_wq)
229 destroy_workqueue(sbi->ll_ra_info.ll_readahead_wq);
230 if (sbi->ll_cache != NULL) {
231 cl_cache_decref(sbi->ll_cache);
232 sbi->ll_cache = NULL;
234 if (sbi->ll_foreign_symlink_prefix) {
235 OBD_FREE(sbi->ll_foreign_symlink_prefix,
236 sbi->ll_foreign_symlink_prefix_size);
237 sbi->ll_foreign_symlink_prefix = NULL;
239 if (sbi->ll_foreign_symlink_upcall) {
240 OBD_FREE(sbi->ll_foreign_symlink_upcall,
241 strlen(sbi->ll_foreign_symlink_upcall) +
243 sbi->ll_foreign_symlink_upcall = NULL;
245 if (sbi->ll_foreign_symlink_upcall_items) {
247 int nb_items = sbi->ll_foreign_symlink_upcall_nb_items;
248 struct ll_foreign_symlink_upcall_item *items =
249 sbi->ll_foreign_symlink_upcall_items;
251 for (i = 0 ; i < nb_items; i++)
252 if (items[i].type == STRING_TYPE)
253 OBD_FREE(items[i].string,
256 OBD_FREE_LARGE(items, nb_items *
257 sizeof(struct ll_foreign_symlink_upcall_item));
258 sbi->ll_foreign_symlink_upcall_items = NULL;
260 ll_free_rw_stats_info(sbi);
261 pcc_super_fini(&sbi->ll_pcc_super);
262 OBD_FREE(sbi, sizeof(*sbi));
267 static int client_common_fill_super(struct super_block *sb, char *md, char *dt)
269 struct inode *root = NULL;
270 struct ll_sb_info *sbi = ll_s2sbi(sb);
271 struct obd_statfs *osfs = NULL;
272 struct ptlrpc_request *request = NULL;
273 struct obd_connect_data *data = NULL;
274 struct obd_uuid *uuid;
275 struct md_op_data *op_data;
276 struct lustre_md lmd;
278 int size, err, checksum;
284 sbi->ll_md_obd = class_name2obd(md);
285 if (!sbi->ll_md_obd) {
286 CERROR("MD %s: not setup or attached\n", md);
300 /* pass client page size via ocd_grant_blkbits, the server should report
301 * back its backend blocksize for grant calculation purpose */
302 data->ocd_grant_blkbits = PAGE_SHIFT;
304 /* indicate MDT features supported by this client */
305 data->ocd_connect_flags = OBD_CONNECT_IBITS | OBD_CONNECT_NODEVOH |
306 OBD_CONNECT_ATTRFID | OBD_CONNECT_GRANT |
307 OBD_CONNECT_VERSION | OBD_CONNECT_BRW_SIZE |
308 OBD_CONNECT_SRVLOCK |
309 OBD_CONNECT_MDS_CAPA | OBD_CONNECT_OSS_CAPA |
310 OBD_CONNECT_CANCELSET | OBD_CONNECT_FID |
311 OBD_CONNECT_AT | OBD_CONNECT_LOV_V3 |
312 OBD_CONNECT_VBR | OBD_CONNECT_FULL20 |
313 OBD_CONNECT_64BITHASH |
314 OBD_CONNECT_EINPROGRESS |
315 OBD_CONNECT_JOBSTATS | OBD_CONNECT_LVB_TYPE |
316 OBD_CONNECT_LAYOUTLOCK | OBD_CONNECT_PINGLESS|
317 OBD_CONNECT_MAX_EASIZE |
318 OBD_CONNECT_FLOCK_DEAD |
319 OBD_CONNECT_DISP_STRIPE | OBD_CONNECT_LFSCK |
320 OBD_CONNECT_OPEN_BY_FID |
321 OBD_CONNECT_DIR_STRIPE |
322 OBD_CONNECT_BULK_MBITS | OBD_CONNECT_CKSUM |
323 OBD_CONNECT_SUBTREE |
324 OBD_CONNECT_MULTIMODRPCS |
325 OBD_CONNECT_GRANT_PARAM |
326 OBD_CONNECT_GRANT_SHRINK |
327 OBD_CONNECT_SHORTIO | OBD_CONNECT_FLAGS2;
329 data->ocd_connect_flags2 = OBD_CONNECT2_DIR_MIGRATE |
330 OBD_CONNECT2_SUM_STATFS |
331 OBD_CONNECT2_OVERSTRIPING |
333 OBD_CONNECT2_LOCK_CONVERT |
334 OBD_CONNECT2_ARCHIVE_ID_ARRAY |
335 OBD_CONNECT2_INC_XID |
337 OBD_CONNECT2_ASYNC_DISCARD |
339 OBD_CONNECT2_CRUSH | OBD_CONNECT2_LSEEK |
340 OBD_CONNECT2_GETATTR_PFID |
341 OBD_CONNECT2_DOM_LVB |
342 OBD_CONNECT2_REP_MBITS |
343 OBD_CONNECT2_ATOMIC_OPEN_LOCK;
345 #ifdef HAVE_LRU_RESIZE_SUPPORT
346 if (test_bit(LL_SBI_LRU_RESIZE, sbi->ll_flags))
347 data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
349 data->ocd_connect_flags |= OBD_CONNECT_ACL_FLAGS;
351 data->ocd_cksum_types = obd_cksum_types_supported_client();
353 if (OBD_FAIL_CHECK(OBD_FAIL_MDC_LIGHTWEIGHT))
354 /* flag mdc connection as lightweight, only used for test
355 * purpose, use with care */
356 data->ocd_connect_flags |= OBD_CONNECT_LIGHTWEIGHT;
358 data->ocd_ibits_known = MDS_INODELOCK_FULL;
359 data->ocd_version = LUSTRE_VERSION_CODE;
361 if (sb->s_flags & SB_RDONLY)
362 data->ocd_connect_flags |= OBD_CONNECT_RDONLY;
363 if (test_bit(LL_SBI_USER_XATTR, sbi->ll_flags))
364 data->ocd_connect_flags |= OBD_CONNECT_XATTR;
367 /* Setting this indicates we correctly support S_NOSEC (See kernel
368 * commit 9e1f1de02c2275d7172e18dc4e7c2065777611bf)
370 sb->s_flags |= SB_NOSEC;
372 sbi->ll_fop = ll_select_file_operations(sbi);
374 /* always ping even if server suppress_pings */
375 if (test_bit(LL_SBI_ALWAYS_PING, sbi->ll_flags))
376 data->ocd_connect_flags &= ~OBD_CONNECT_PINGLESS;
378 obd_connect_set_secctx(data);
379 if (ll_sbi_has_encrypt(sbi)) {
380 obd_connect_set_name_enc(data);
381 obd_connect_set_enc(data);
384 #if defined(CONFIG_SECURITY)
385 data->ocd_connect_flags2 |= OBD_CONNECT2_SELINUX_POLICY;
388 data->ocd_brw_size = MD_MAX_BRW_SIZE;
390 err = obd_connect(NULL, &sbi->ll_md_exp, sbi->ll_md_obd,
391 &sbi->ll_sb_uuid, data, sbi->ll_cache);
393 LCONSOLE_ERROR_MSG(0x14f, "An MDT (md %s) is performing "
394 "recovery, of which this client is not a "
395 "part. Please wait for recovery to complete,"
396 " abort, or time out.\n", md);
399 CERROR("cannot connect to %s: rc = %d\n", md, err);
403 sbi->ll_md_exp->exp_connect_data = *data;
405 err = obd_fid_init(sbi->ll_md_exp->exp_obd, sbi->ll_md_exp,
406 LUSTRE_SEQ_METADATA);
408 CERROR("%s: Can't init metadata layer FID infrastructure, "
409 "rc = %d\n", sbi->ll_md_exp->exp_obd->obd_name, err);
413 /* For mount, we only need fs info from MDT0, and also in DNE, it
414 * can make sure the client can be mounted as long as MDT0 is
416 err = obd_statfs(NULL, sbi->ll_md_exp, osfs,
417 ktime_get_seconds() - sbi->ll_statfs_max_age,
418 OBD_STATFS_FOR_MDT0);
420 GOTO(out_md_fid, err);
422 /* This needs to be after statfs to ensure connect has finished.
423 * Note that "data" does NOT contain the valid connect reply.
424 * If connecting to a 1.8 server there will be no LMV device, so
425 * we can access the MDC export directly and exp_connect_flags will
426 * be non-zero, but if accessing an upgraded 2.1 server it will
427 * have the correct flags filled in.
428 * XXX: fill in the LMV exp_connect_flags from MDC(s). */
429 valid = exp_connect_flags(sbi->ll_md_exp) & CLIENT_CONNECT_MDT_REQD;
430 if (exp_connect_flags(sbi->ll_md_exp) != 0 &&
431 valid != CLIENT_CONNECT_MDT_REQD) {
434 OBD_ALLOC_WAIT(buf, PAGE_SIZE);
435 obd_connect_flags2str(buf, PAGE_SIZE,
436 valid ^ CLIENT_CONNECT_MDT_REQD, 0, ",");
437 LCONSOLE_ERROR_MSG(0x170, "Server %s does not support "
438 "feature(s) needed for correct operation "
439 "of this client (%s). Please upgrade "
440 "server or downgrade client.\n",
441 sbi->ll_md_exp->exp_obd->obd_name, buf);
442 OBD_FREE(buf, PAGE_SIZE);
443 GOTO(out_md_fid, err = -EPROTO);
446 size = sizeof(*data);
447 err = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_CONN_DATA),
448 KEY_CONN_DATA, &size, data);
450 CERROR("%s: Get connect data failed: rc = %d\n",
451 sbi->ll_md_exp->exp_obd->obd_name, err);
452 GOTO(out_md_fid, err);
455 LASSERT(osfs->os_bsize);
456 sb->s_blocksize = osfs->os_bsize;
457 sb->s_blocksize_bits = log2(osfs->os_bsize);
458 sb->s_magic = LL_SUPER_MAGIC;
459 sb->s_maxbytes = MAX_LFS_FILESIZE;
460 sbi->ll_inode_cache_enabled = 1;
461 sbi->ll_namelen = osfs->os_namelen;
462 sbi->ll_mnt.mnt = current->fs->root.mnt;
463 sbi->ll_mnt_ns = current->nsproxy->mnt_ns;
465 if (test_bit(LL_SBI_USER_XATTR, sbi->ll_flags) &&
466 !(data->ocd_connect_flags & OBD_CONNECT_XATTR)) {
467 LCONSOLE_INFO("Disabling user_xattr feature because "
468 "it is not supported on the server\n");
469 clear_bit(LL_SBI_USER_XATTR, sbi->ll_flags);
472 if (data->ocd_connect_flags & OBD_CONNECT_ACL) {
474 sb->s_flags |= SB_POSIXACL;
476 set_bit(LL_SBI_ACL, sbi->ll_flags);
478 LCONSOLE_INFO("client wants to enable acl, but mdt not!\n");
480 sb->s_flags &= ~SB_POSIXACL;
482 clear_bit(LL_SBI_ACL, sbi->ll_flags);
485 if (data->ocd_connect_flags & OBD_CONNECT_64BITHASH)
486 set_bit(LL_SBI_64BIT_HASH, sbi->ll_flags);
488 if (data->ocd_connect_flags & OBD_CONNECT_LAYOUTLOCK)
489 set_bit(LL_SBI_LAYOUT_LOCK, sbi->ll_flags);
491 if (obd_connect_has_secctx(data))
492 set_bit(LL_SBI_FILE_SECCTX, sbi->ll_flags);
494 if (ll_sbi_has_encrypt(sbi) && !obd_connect_has_enc(data)) {
495 if (ll_sb_has_test_dummy_encryption(sb))
496 LCONSOLE_WARN("%s: server %s does not support encryption feature, encryption deactivated.\n",
498 sbi->ll_md_exp->exp_obd->obd_name);
499 ll_sbi_set_encrypt(sbi, false);
502 if (ll_sbi_has_name_encrypt(sbi) && !obd_connect_has_name_enc(data)) {
503 struct lustre_sb_info *lsi = s2lsi(sb);
505 if (ll_sb_has_test_dummy_encryption(sb))
506 LCONSOLE_WARN("%s: server %s does not support name encryption, not using it.\n",
508 sbi->ll_md_exp->exp_obd->obd_name);
509 lsi->lsi_flags &= ~LSI_FILENAME_ENC;
510 ll_sbi_set_name_encrypt(sbi, false);
513 if (data->ocd_ibits_known & MDS_INODELOCK_XATTR) {
514 if (!(data->ocd_connect_flags & OBD_CONNECT_MAX_EASIZE)) {
515 LCONSOLE_INFO("%s: disabling xattr cache due to "
516 "unknown maximum xattr size.\n", dt);
517 } else if (!sbi->ll_xattr_cache_set) {
518 /* If xattr_cache is already set (no matter 0 or 1)
519 * during processing llog, it won't be enabled here. */
520 set_bit(LL_SBI_XATTR_CACHE, sbi->ll_flags);
521 sbi->ll_xattr_cache_enabled = 1;
525 sbi->ll_dt_obd = class_name2obd(dt);
526 if (!sbi->ll_dt_obd) {
527 CERROR("DT %s: not setup or attached\n", dt);
528 GOTO(out_md_fid, err = -ENODEV);
531 /* pass client page size via ocd_grant_blkbits, the server should report
532 * back its backend blocksize for grant calculation purpose */
533 data->ocd_grant_blkbits = PAGE_SHIFT;
535 /* indicate OST features supported by this client */
536 data->ocd_connect_flags = OBD_CONNECT_GRANT | OBD_CONNECT_VERSION |
537 OBD_CONNECT_REQPORTAL | OBD_CONNECT_BRW_SIZE |
538 OBD_CONNECT_CANCELSET | OBD_CONNECT_FID |
539 OBD_CONNECT_SRVLOCK |
540 OBD_CONNECT_AT | OBD_CONNECT_OSS_CAPA |
541 OBD_CONNECT_VBR | OBD_CONNECT_FULL20 |
542 OBD_CONNECT_64BITHASH | OBD_CONNECT_MAXBYTES |
543 OBD_CONNECT_EINPROGRESS |
544 OBD_CONNECT_JOBSTATS | OBD_CONNECT_LVB_TYPE |
545 OBD_CONNECT_LAYOUTLOCK |
546 OBD_CONNECT_PINGLESS | OBD_CONNECT_LFSCK |
547 OBD_CONNECT_BULK_MBITS | OBD_CONNECT_SHORTIO |
548 OBD_CONNECT_FLAGS2 | OBD_CONNECT_GRANT_SHRINK;
549 data->ocd_connect_flags2 = OBD_CONNECT2_LOCKAHEAD |
550 OBD_CONNECT2_INC_XID | OBD_CONNECT2_LSEEK |
551 OBD_CONNECT2_REP_MBITS;
553 if (!OBD_FAIL_CHECK(OBD_FAIL_OSC_CONNECT_GRANT_PARAM))
554 data->ocd_connect_flags |= OBD_CONNECT_GRANT_PARAM;
556 /* OBD_CONNECT_CKSUM should always be set, even if checksums are
557 * disabled by default, because it can still be enabled on the
558 * fly via /sys. As a consequence, we still need to come to an
559 * agreement on the supported algorithms at connect time
561 data->ocd_connect_flags |= OBD_CONNECT_CKSUM;
563 if (OBD_FAIL_CHECK(OBD_FAIL_OSC_CKSUM_ADLER_ONLY))
564 data->ocd_cksum_types = OBD_CKSUM_ADLER;
566 data->ocd_cksum_types = obd_cksum_types_supported_client();
568 #ifdef HAVE_LRU_RESIZE_SUPPORT
569 data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
571 /* always ping even if server suppress_pings */
572 if (test_bit(LL_SBI_ALWAYS_PING, sbi->ll_flags))
573 data->ocd_connect_flags &= ~OBD_CONNECT_PINGLESS;
575 if (ll_sbi_has_encrypt(sbi))
576 obd_connect_set_enc(data);
578 CDEBUG(D_RPCTRACE, "ocd_connect_flags: %#llx ocd_version: %d "
579 "ocd_grant: %d\n", data->ocd_connect_flags,
580 data->ocd_version, data->ocd_grant);
582 sbi->ll_dt_obd->obd_upcall.onu_owner = &sbi->ll_lco;
583 sbi->ll_dt_obd->obd_upcall.onu_upcall = cl_ocd_update;
585 data->ocd_brw_size = DT_MAX_BRW_SIZE;
587 err = obd_connect(NULL, &sbi->ll_dt_exp, sbi->ll_dt_obd,
588 &sbi->ll_sb_uuid, data, sbi->ll_cache);
590 LCONSOLE_ERROR_MSG(0x150, "An OST (dt %s) is performing "
591 "recovery, of which this client is not a "
592 "part. Please wait for recovery to "
593 "complete, abort, or time out.\n", dt);
596 CERROR("%s: Cannot connect to %s: rc = %d\n",
597 sbi->ll_dt_exp->exp_obd->obd_name, dt, err);
601 if (ll_sbi_has_encrypt(sbi) &&
602 !obd_connect_has_enc(&sbi->ll_dt_obd->u.lov.lov_ocd)) {
603 if (ll_sb_has_test_dummy_encryption(sb))
604 LCONSOLE_WARN("%s: server %s does not support encryption feature, encryption deactivated.\n",
606 ll_sbi_set_encrypt(sbi, false);
607 } else if (ll_sb_has_test_dummy_encryption(sb)) {
608 LCONSOLE_WARN("Test dummy encryption mode enabled\n");
611 sbi->ll_dt_exp->exp_connect_data = *data;
613 /* Don't change value if it was specified in the config log */
614 if (sbi->ll_ra_info.ra_max_read_ahead_whole_pages == -1) {
615 sbi->ll_ra_info.ra_max_read_ahead_whole_pages =
616 max_t(unsigned long, SBI_DEFAULT_READ_AHEAD_WHOLE_MAX,
617 (data->ocd_brw_size >> PAGE_SHIFT));
618 if (sbi->ll_ra_info.ra_max_read_ahead_whole_pages >
619 sbi->ll_ra_info.ra_max_pages_per_file)
620 sbi->ll_ra_info.ra_max_read_ahead_whole_pages =
621 sbi->ll_ra_info.ra_max_pages_per_file;
624 err = obd_fid_init(sbi->ll_dt_exp->exp_obd, sbi->ll_dt_exp,
625 LUSTRE_SEQ_METADATA);
627 CERROR("%s: Can't init data layer FID infrastructure, "
628 "rc = %d\n", sbi->ll_dt_exp->exp_obd->obd_name, err);
632 mutex_lock(&sbi->ll_lco.lco_lock);
633 sbi->ll_lco.lco_flags = data->ocd_connect_flags;
634 sbi->ll_lco.lco_md_exp = sbi->ll_md_exp;
635 sbi->ll_lco.lco_dt_exp = sbi->ll_dt_exp;
636 mutex_unlock(&sbi->ll_lco.lco_lock);
638 fid_zero(&sbi->ll_root_fid);
639 err = md_get_root(sbi->ll_md_exp, get_mount_fileset(sb),
642 CERROR("cannot mds_connect: rc = %d\n", err);
643 GOTO(out_lock_cn_cb, err);
645 if (!fid_is_sane(&sbi->ll_root_fid)) {
646 CERROR("%s: Invalid root fid "DFID" during mount\n",
647 sbi->ll_md_exp->exp_obd->obd_name,
648 PFID(&sbi->ll_root_fid));
649 GOTO(out_lock_cn_cb, err = -EINVAL);
651 CDEBUG(D_SUPER, "rootfid "DFID"\n", PFID(&sbi->ll_root_fid));
653 sb->s_op = &lustre_super_operations;
654 sb->s_xattr = ll_xattr_handlers;
655 #if THREAD_SIZE >= 8192 /*b=17630*/
656 sb->s_export_op = &lustre_export_operations;
658 #ifdef HAVE_LUSTRE_CRYPTO
659 llcrypt_set_ops(sb, &lustre_cryptops);
663 * XXX: move this to after cbd setup? */
664 valid = OBD_MD_FLGETATTR | OBD_MD_FLBLOCKS | OBD_MD_FLMODEASIZE |
666 if (test_bit(LL_SBI_ACL, sbi->ll_flags))
667 valid |= OBD_MD_FLACL;
669 OBD_ALLOC_PTR(op_data);
671 GOTO(out_lock_cn_cb, err = -ENOMEM);
673 op_data->op_fid1 = sbi->ll_root_fid;
674 op_data->op_mode = 0;
675 op_data->op_valid = valid;
677 err = md_getattr(sbi->ll_md_exp, op_data, &request);
679 /* We need enc ctx info, so reset it in op_data to
680 * prevent it from being freed.
682 encctx = op_data->op_file_encctx;
683 encctxlen = op_data->op_file_encctx_size;
684 op_data->op_file_encctx = NULL;
685 op_data->op_file_encctx_size = 0;
686 OBD_FREE_PTR(op_data);
688 CERROR("%s: md_getattr failed for root: rc = %d\n",
689 sbi->ll_md_exp->exp_obd->obd_name, err);
690 GOTO(out_lock_cn_cb, err);
693 err = md_get_lustre_md(sbi->ll_md_exp, &request->rq_pill,
694 sbi->ll_dt_exp, sbi->ll_md_exp, &lmd);
696 CERROR("failed to understand root inode md: rc = %d\n", err);
697 ptlrpc_req_finished(request);
698 GOTO(out_lock_cn_cb, err);
701 LASSERT(fid_is_sane(&sbi->ll_root_fid));
702 api32 = test_bit(LL_SBI_32BIT_API, sbi->ll_flags);
703 root = ll_iget(sb, cl_fid_build_ino(&sbi->ll_root_fid, api32), &lmd);
704 md_free_lustre_md(sbi->ll_md_exp, &lmd);
708 err = IS_ERR(root) ? PTR_ERR(root) : -EBADF;
710 CERROR("%s: bad ll_iget() for root: rc = %d\n",
711 sbi->ll_fsname, err);
712 ptlrpc_req_finished(request);
718 "server returned encryption ctx for root inode "DFID"\n",
719 PFID(&sbi->ll_root_fid));
720 err = ll_set_encflags(root, encctx, encctxlen, true);
722 CWARN("%s: cannot set enc ctx for "DFID": rc = %d\n",
724 PFID(&sbi->ll_root_fid), err);
726 ptlrpc_req_finished(request);
728 checksum = test_bit(LL_SBI_CHECKSUM, sbi->ll_flags);
729 if (sbi->ll_checksum_set) {
730 err = obd_set_info_async(NULL, sbi->ll_dt_exp,
731 sizeof(KEY_CHECKSUM), KEY_CHECKSUM,
732 sizeof(checksum), &checksum, NULL);
734 CERROR("%s: Set checksum failed: rc = %d\n",
735 sbi->ll_dt_exp->exp_obd->obd_name, err);
741 sb->s_root = d_make_root(root);
742 if (sb->s_root == NULL) {
744 CERROR("%s: can't make root dentry: rc = %d\n",
745 sbi->ll_fsname, err);
749 sbi->ll_sdev_orig = sb->s_dev;
751 /* We set sb->s_dev equal on all lustre clients in order to support
752 * NFS export clustering. NFSD requires that the FSID be the same
754 /* s_dev is also used in lt_compare() to compare two fs, but that is
755 * only a node-local comparison. */
756 uuid = obd_get_uuid(sbi->ll_md_exp);
758 sb->s_dev = get_uuid2int(uuid->uuid, strlen(uuid->uuid));
765 if (sbi->ll_dt_obd) {
766 err = sysfs_create_link(&sbi->ll_kset.kobj,
767 &sbi->ll_dt_obd->obd_kset.kobj,
768 sbi->ll_dt_obd->obd_type->typ_name);
770 CERROR("%s: could not register %s in llite: rc = %d\n",
771 dt, sbi->ll_fsname, err);
776 if (sbi->ll_md_obd) {
777 err = sysfs_create_link(&sbi->ll_kset.kobj,
778 &sbi->ll_md_obd->obd_kset.kobj,
779 sbi->ll_md_obd->obd_type->typ_name);
781 CERROR("%s: could not register %s in llite: rc = %d\n",
782 md, sbi->ll_fsname, err);
791 obd_fid_fini(sbi->ll_dt_exp->exp_obd);
793 obd_disconnect(sbi->ll_dt_exp);
794 sbi->ll_dt_exp = NULL;
795 sbi->ll_dt_obd = NULL;
797 obd_fid_fini(sbi->ll_md_exp->exp_obd);
799 obd_disconnect(sbi->ll_md_exp);
800 sbi->ll_md_exp = NULL;
801 sbi->ll_md_obd = NULL;
810 int ll_get_max_mdsize(struct ll_sb_info *sbi, int *lmmsize)
814 size = sizeof(*lmmsize);
815 rc = obd_get_info(NULL, sbi->ll_dt_exp, sizeof(KEY_MAX_EASIZE),
816 KEY_MAX_EASIZE, &size, lmmsize);
818 CERROR("%s: cannot get max LOV EA size: rc = %d\n",
819 sbi->ll_dt_exp->exp_obd->obd_name, rc);
823 CDEBUG(D_INFO, "max LOV ea size: %d\n", *lmmsize);
826 rc = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_MAX_EASIZE),
827 KEY_MAX_EASIZE, &size, lmmsize);
829 CERROR("Get max mdsize error rc %d\n", rc);
831 CDEBUG(D_INFO, "max LMV ea size: %d\n", *lmmsize);
837 * Get the value of the default_easize parameter.
839 * \see client_obd::cl_default_mds_easize
841 * \param[in] sbi superblock info for this filesystem
842 * \param[out] lmmsize pointer to storage location for value
844 * \retval 0 on success
845 * \retval negative negated errno on failure
847 int ll_get_default_mdsize(struct ll_sb_info *sbi, int *lmmsize)
852 rc = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_DEFAULT_EASIZE),
853 KEY_DEFAULT_EASIZE, &size, lmmsize);
855 CERROR("Get default mdsize error rc %d\n", rc);
861 * Set the default_easize parameter to the given value.
863 * \see client_obd::cl_default_mds_easize
865 * \param[in] sbi superblock info for this filesystem
866 * \param[in] lmmsize the size to set
868 * \retval 0 on success
869 * \retval negative negated errno on failure
871 int ll_set_default_mdsize(struct ll_sb_info *sbi, int lmmsize)
875 if (lmmsize < sizeof(struct lov_mds_md) ||
876 lmmsize > OBD_MAX_DEFAULT_EA_SIZE)
879 rc = obd_set_info_async(NULL, sbi->ll_md_exp,
880 sizeof(KEY_DEFAULT_EASIZE), KEY_DEFAULT_EASIZE,
881 sizeof(int), &lmmsize, NULL);
886 static void client_common_put_super(struct super_block *sb)
888 struct ll_sb_info *sbi = ll_s2sbi(sb);
893 obd_fid_fini(sbi->ll_dt_exp->exp_obd);
894 obd_disconnect(sbi->ll_dt_exp);
895 sbi->ll_dt_exp = NULL;
897 ll_debugfs_unregister_super(sb);
899 obd_fid_fini(sbi->ll_md_exp->exp_obd);
900 obd_disconnect(sbi->ll_md_exp);
901 sbi->ll_md_exp = NULL;
906 void ll_kill_super(struct super_block *sb)
908 struct ll_sb_info *sbi;
912 if (!(sb->s_flags & SB_ACTIVE))
916 /* we need restore s_dev from changed for clustred NFS before put_super
917 * because new kernels have cached s_dev and change sb->s_dev in
918 * put_super not affected real removing devices */
920 sb->s_dev = sbi->ll_sdev_orig;
922 /* wait running statahead threads to quit */
923 while (atomic_read(&sbi->ll_sa_running) > 0)
924 schedule_timeout_uninterruptible(
925 cfs_time_seconds(1) >> 3);
931 /* Since we use this table for ll_sbi_flags_seq_show make
932 * sure what you want displayed for a specific token that
933 * is listed more than once below be listed first. For
934 * example we want "checksum" displayed, not "nochecksum"
937 static const match_table_t ll_sbi_flags_name = {
938 {LL_SBI_NOLCK, "nolock"},
939 {LL_SBI_CHECKSUM, "checksum"},
940 {LL_SBI_CHECKSUM, "nochecksum"},
941 {LL_SBI_LOCALFLOCK, "localflock"},
942 {LL_SBI_FLOCK, "flock"},
943 {LL_SBI_FLOCK, "noflock"},
944 {LL_SBI_USER_XATTR, "user_xattr"},
945 {LL_SBI_USER_XATTR, "nouser_xattr"},
946 {LL_SBI_LRU_RESIZE, "lruresize"},
947 {LL_SBI_LRU_RESIZE, "nolruresize"},
948 {LL_SBI_LAZYSTATFS, "lazystatfs"},
949 {LL_SBI_LAZYSTATFS, "nolazystatfs"},
950 {LL_SBI_32BIT_API, "32bitapi"},
951 {LL_SBI_USER_FID2PATH, "user_fid2path"},
952 {LL_SBI_USER_FID2PATH, "nouser_fid2path"},
953 {LL_SBI_VERBOSE, "verbose"},
954 {LL_SBI_VERBOSE, "noverbose"},
955 {LL_SBI_ALWAYS_PING, "always_ping"},
956 {LL_SBI_TEST_DUMMY_ENCRYPTION, "test_dummy_encryption=%s"},
957 {LL_SBI_TEST_DUMMY_ENCRYPTION, "test_dummy_encryption"},
958 {LL_SBI_ENCRYPT, "encrypt"},
959 {LL_SBI_ENCRYPT, "noencrypt"},
960 {LL_SBI_FOREIGN_SYMLINK, "foreign_symlink=%s"},
961 {LL_SBI_NUM_MOUNT_OPT, NULL},
964 {LL_SBI_AGL_ENABLED, "agl"},
965 {LL_SBI_64BIT_HASH, "64bit_hash"},
966 {LL_SBI_LAYOUT_LOCK, "layout"},
967 {LL_SBI_XATTR_CACHE, "xattr_cache"},
968 {LL_SBI_NOROOTSQUASH, "norootsquash"},
969 {LL_SBI_FAST_READ, "fast_read"},
970 {LL_SBI_FILE_SECCTX, "file_secctx"},
971 {LL_SBI_TINY_WRITE, "tiny_write"},
972 {LL_SBI_FILE_HEAT, "file_heat"},
973 {LL_SBI_PARALLEL_DIO, "parallel_dio"},
974 {LL_SBI_ENCRYPT_NAME, "name_encrypt"},
977 int ll_sbi_flags_seq_show(struct seq_file *m, void *v)
979 struct super_block *sb = m->private;
982 for (i = 0; i < LL_SBI_NUM_FLAGS; i++) {
985 if (!test_bit(i, ll_s2sbi(sb)->ll_flags))
988 for (j = 0; j < ARRAY_SIZE(ll_sbi_flags_name); j++) {
989 if (ll_sbi_flags_name[j].token == i &&
990 ll_sbi_flags_name[j].pattern) {
992 ll_sbi_flags_name[j].pattern);
1001 /* non-client-specific mount options are parsed in lmd_parse */
1002 static int ll_options(char *options, struct super_block *sb)
1004 struct ll_sb_info *sbi = ll_s2sbi(sb);
1005 char *s2, *s1, *opts;
1012 /* Don't stomp on lmd_opts */
1013 opts = kstrdup(options, GFP_KERNEL);
1019 CDEBUG(D_CONFIG, "Parsing opts %s\n", options);
1021 while ((s1 = strsep(&opts, ",")) != NULL) {
1022 substring_t args[MAX_OPT_ARGS];
1023 bool turn_off = false;
1029 CDEBUG(D_SUPER, "next opt=%s\n", s1);
1031 if (strncmp(s1, "no", 2) == 0)
1035 * Initialize args struct so we know whether arg was
1036 * found; some options take optional arguments.
1039 args[0].from = NULL;
1040 token = match_token(s1, ll_sbi_flags_name, args);
1041 if (token == LL_SBI_NUM_MOUNT_OPT) {
1042 if (match_wildcard("context", s1) ||
1043 match_wildcard("fscontext", s1) ||
1044 match_wildcard("defcontext", s1) ||
1045 match_wildcard("rootcontext",s1))
1048 LCONSOLE_ERROR_MSG(0x152,
1049 "Unknown option '%s', won't mount.\n",
1056 case LL_SBI_32BIT_API:
1057 case LL_SBI_64BIT_HASH:
1058 case LL_SBI_ALWAYS_PING:
1059 set_bit(token, sbi->ll_flags);
1063 clear_bit(LL_SBI_LOCALFLOCK, sbi->ll_flags);
1065 clear_bit(LL_SBI_FLOCK, sbi->ll_flags);
1067 set_bit(token, sbi->ll_flags);
1070 case LL_SBI_LOCALFLOCK:
1071 clear_bit(LL_SBI_FLOCK, sbi->ll_flags);
1072 set_bit(token, sbi->ll_flags);
1075 case LL_SBI_CHECKSUM:
1076 sbi->ll_checksum_set = 1;
1078 case LL_SBI_USER_XATTR:
1079 case LL_SBI_USER_FID2PATH:
1080 case LL_SBI_LRU_RESIZE:
1081 case LL_SBI_LAZYSTATFS:
1082 case LL_SBI_VERBOSE:
1084 clear_bit(token, sbi->ll_flags);
1086 set_bit(token, sbi->ll_flags);
1088 case LL_SBI_TEST_DUMMY_ENCRYPTION: {
1089 #ifdef HAVE_LUSTRE_CRYPTO
1090 #ifdef HAVE_FSCRYPT_DUMMY_CONTEXT_ENABLED
1091 set_bit(token, sbi->ll_flags);
1093 struct lustre_sb_info *lsi = s2lsi(sb);
1095 err = llcrypt_set_test_dummy_encryption(sb, &args[0],
1096 &lsi->lsi_dummy_enc_ctx);
1102 "Can't change test_dummy_encryption");
1103 else if (err == -EINVAL)
1105 "Value of option \"%s\" unrecognized",
1109 "Error processing option \"%s\" [%d]",
1114 LCONSOLE_WARN("Test dummy encryption mount option ignored: encryption not supported\n");
1118 case LL_SBI_ENCRYPT:
1119 #ifdef HAVE_LUSTRE_CRYPTO
1121 clear_bit(token, sbi->ll_flags);
1123 set_bit(token, sbi->ll_flags);
1125 LCONSOLE_WARN("noencrypt or encrypt mount option ignored: encryption not supported\n");
1128 case LL_SBI_FOREIGN_SYMLINK:
1129 /* non-default prefix provided ? */
1134 /* path must be absolute */
1135 if (args->from[0] != '/') {
1136 LCONSOLE_ERROR_MSG(0x152,
1137 "foreign prefix '%s' must be an absolute path\n",
1142 old_len = sbi->ll_foreign_symlink_prefix_size;
1143 old = sbi->ll_foreign_symlink_prefix;
1144 /* alloc for path length and '\0' */
1145 sbi->ll_foreign_symlink_prefix = match_strdup(args);
1146 if (!sbi->ll_foreign_symlink_prefix) {
1147 /* restore previous */
1148 sbi->ll_foreign_symlink_prefix = old;
1149 sbi->ll_foreign_symlink_prefix_size =
1153 sbi->ll_foreign_symlink_prefix_size =
1154 args->to - args->from + 1;
1155 OBD_ALLOC_POST(sbi->ll_foreign_symlink_prefix,
1156 sbi->ll_foreign_symlink_prefix_size,
1159 OBD_FREE(old, old_len);
1161 /* enable foreign symlink support */
1162 set_bit(token, sbi->ll_flags);
1164 LCONSOLE_ERROR_MSG(0x152,
1165 "invalid %s option\n", s1);
1176 void ll_lli_init(struct ll_inode_info *lli)
1178 lli->lli_inode_magic = LLI_INODE_MAGIC;
1180 rwlock_init(&lli->lli_lock);
1181 lli->lli_posix_acl = NULL;
1182 /* Do not set lli_fid, it has been initialized already. */
1183 fid_zero(&lli->lli_pfid);
1184 lli->lli_mds_read_och = NULL;
1185 lli->lli_mds_write_och = NULL;
1186 lli->lli_mds_exec_och = NULL;
1187 lli->lli_open_fd_read_count = 0;
1188 lli->lli_open_fd_write_count = 0;
1189 lli->lli_open_fd_exec_count = 0;
1190 mutex_init(&lli->lli_och_mutex);
1191 spin_lock_init(&lli->lli_agl_lock);
1192 spin_lock_init(&lli->lli_layout_lock);
1193 ll_layout_version_set(lli, CL_LAYOUT_GEN_NONE);
1194 lli->lli_clob = NULL;
1196 init_rwsem(&lli->lli_xattrs_list_rwsem);
1197 mutex_init(&lli->lli_xattrs_enq_lock);
1199 LASSERT(lli->lli_vfs_inode.i_mode != 0);
1200 if (S_ISDIR(lli->lli_vfs_inode.i_mode)) {
1201 lli->lli_opendir_key = NULL;
1202 lli->lli_sai = NULL;
1203 spin_lock_init(&lli->lli_sa_lock);
1204 lli->lli_opendir_pid = 0;
1205 lli->lli_sa_enabled = 0;
1206 init_rwsem(&lli->lli_lsm_sem);
1208 mutex_init(&lli->lli_size_mutex);
1209 mutex_init(&lli->lli_setattr_mutex);
1210 lli->lli_symlink_name = NULL;
1211 ll_trunc_sem_init(&lli->lli_trunc_sem);
1212 range_lock_tree_init(&lli->lli_write_tree);
1213 init_rwsem(&lli->lli_glimpse_sem);
1214 lli->lli_glimpse_time = ktime_set(0, 0);
1215 INIT_LIST_HEAD(&lli->lli_agl_list);
1216 lli->lli_agl_index = 0;
1217 lli->lli_async_rc = 0;
1218 spin_lock_init(&lli->lli_heat_lock);
1219 obd_heat_clear(lli->lli_heat_instances, OBD_HEAT_COUNT);
1220 lli->lli_heat_flags = 0;
1221 mutex_init(&lli->lli_pcc_lock);
1222 lli->lli_pcc_state = PCC_STATE_FL_NONE;
1223 lli->lli_pcc_inode = NULL;
1224 lli->lli_pcc_dsflags = PCC_DATASET_INVALID;
1225 lli->lli_pcc_generation = 0;
1226 mutex_init(&lli->lli_group_mutex);
1227 lli->lli_group_users = 0;
1228 lli->lli_group_gid = 0;
1230 mutex_init(&lli->lli_layout_mutex);
1231 memset(lli->lli_jobid, 0, sizeof(lli->lli_jobid));
1232 /* ll_cl_context initialize */
1233 INIT_LIST_HEAD(&lli->lli_lccs);
1236 #define MAX_STRING_SIZE 128
1238 #ifndef HAVE_SUPER_SETUP_BDI_NAME
1239 #ifndef HAVE_BDI_CAP_MAP_COPY
1240 # define BDI_CAP_MAP_COPY 0
1243 static int super_setup_bdi_name(struct super_block *sb, char *fmt, ...)
1245 struct lustre_sb_info *lsi = s2lsi(sb);
1246 char buf[MAX_STRING_SIZE];
1250 err = bdi_init(&lsi->lsi_bdi);
1254 lsi->lsi_flags |= LSI_BDI_INITIALIZED;
1255 lsi->lsi_bdi.capabilities = BDI_CAP_MAP_COPY;
1256 lsi->lsi_bdi.name = "lustre";
1257 va_start(args, fmt);
1258 vsnprintf(buf, MAX_STRING_SIZE, fmt, args);
1260 err = bdi_register(&lsi->lsi_bdi, NULL, "%s", buf);
1263 sb->s_bdi = &lsi->lsi_bdi;
1267 #endif /* !HAVE_SUPER_SETUP_BDI_NAME */
1269 int ll_fill_super(struct super_block *sb)
1271 struct lustre_profile *lprof = NULL;
1272 struct lustre_sb_info *lsi = s2lsi(sb);
1273 struct ll_sb_info *sbi = NULL;
1274 char *dt = NULL, *md = NULL;
1275 char *profilenm = get_profile_name(sb);
1276 struct config_llog_instance *cfg;
1277 /* %p for void* in printf needs 16+2 characters: 0xffffffffffffffff */
1278 const int instlen = LUSTRE_MAXINSTANCE + 2;
1279 unsigned long cfg_instance = ll_get_cfg_instance(sb);
1280 char name[MAX_STRING_SIZE];
1289 /* for ASLR, to map between cfg_instance and hashed ptr */
1290 CDEBUG(D_VFSTRACE, "VFS Op: cfg_instance %s-%016lx (sb %p)\n",
1291 profilenm, cfg_instance, sb);
1293 OBD_RACE(OBD_FAIL_LLITE_RACE_MOUNT);
1297 GOTO(out_free_cfg, err = -ENOMEM);
1299 /* client additional sb info */
1300 lsi->lsi_llsbi = sbi = ll_init_sbi();
1302 GOTO(out_free_cfg, err = PTR_ERR(sbi));
1304 err = ll_options(lsi->lsi_lmd->lmd_opts, sb);
1306 GOTO(out_free_cfg, err);
1308 if (ll_sb_has_test_dummy_encryption(sb))
1309 /* enable filename encryption by default for dummy enc mode */
1310 lsi->lsi_flags |= LSI_FILENAME_ENC;
1312 /* filename encryption is disabled by default */
1313 lsi->lsi_flags &= ~LSI_FILENAME_ENC;
1315 /* kernel >= 2.6.38 store dentry operations in sb->s_d_op. */
1316 sb->s_d_op = &ll_d_ops;
1319 generate_random_uuid(uuid.b);
1320 snprintf(sbi->ll_sb_uuid.uuid, sizeof(sbi->ll_sb_uuid), "%pU", uuid.b);
1322 CDEBUG(D_CONFIG, "llite sb uuid: %s\n", sbi->ll_sb_uuid.uuid);
1325 len = strlen(profilenm);
1326 ptr = strrchr(profilenm, '-');
1327 if (ptr && (strcmp(ptr, "-client") == 0))
1330 if (len > LUSTRE_MAXFSNAME) {
1331 if (unlikely(len >= MAX_STRING_SIZE))
1332 len = MAX_STRING_SIZE - 1;
1333 strncpy(name, profilenm, len);
1335 err = -ENAMETOOLONG;
1336 CERROR("%s: fsname longer than %u characters: rc = %d\n",
1337 name, LUSTRE_MAXFSNAME, err);
1338 GOTO(out_free_cfg, err);
1340 strncpy(sbi->ll_fsname, profilenm, len);
1341 sbi->ll_fsname[len] = '\0';
1344 snprintf(name, sizeof(name), "%.*s-%016lx", len,
1345 profilenm, cfg_instance);
1347 err = super_setup_bdi_name(sb, "%s", name);
1349 GOTO(out_free_cfg, err);
1351 /* disable kernel readahead */
1352 sb->s_bdi->ra_pages = 0;
1353 #ifdef HAVE_BDI_IO_PAGES
1354 sb->s_bdi->io_pages = 0;
1357 /* Call ll_debugfs_register_super() before lustre_process_log()
1358 * so that "llite.*.*" params can be processed correctly.
1360 err = ll_debugfs_register_super(sb, name);
1362 CERROR("%s: could not register mountpoint in llite: rc = %d\n",
1363 sbi->ll_fsname, err);
1367 /* The cfg_instance is a value unique to this super, in case some
1368 * joker tries to mount the same fs at two mount points.
1370 cfg->cfg_instance = cfg_instance;
1371 cfg->cfg_uuid = lsi->lsi_llsbi->ll_sb_uuid;
1372 cfg->cfg_callback = class_config_llog_handler;
1373 cfg->cfg_sub_clds = CONFIG_SUB_CLIENT;
1374 /* set up client obds */
1375 err = lustre_process_log(sb, profilenm, cfg);
1377 GOTO(out_debugfs, err);
1379 /* Profile set with LCFG_MOUNTOPT so we can find our mdc and osc obds */
1380 lprof = class_get_profile(profilenm);
1381 if (lprof == NULL) {
1382 LCONSOLE_ERROR_MSG(0x156, "The client profile '%s' could not be"
1383 " read from the MGS. Does that filesystem "
1384 "exist?\n", profilenm);
1385 GOTO(out_debugfs, err = -EINVAL);
1387 CDEBUG(D_CONFIG, "Found profile %s: mdc=%s osc=%s\n", profilenm,
1388 lprof->lp_md, lprof->lp_dt);
1390 dt_len = strlen(lprof->lp_dt) + instlen + 2;
1391 OBD_ALLOC(dt, dt_len);
1393 GOTO(out_profile, err = -ENOMEM);
1394 snprintf(dt, dt_len - 1, "%s-%016lx", lprof->lp_dt, cfg_instance);
1396 md_len = strlen(lprof->lp_md) + instlen + 2;
1397 OBD_ALLOC(md, md_len);
1399 GOTO(out_free_dt, err = -ENOMEM);
1400 snprintf(md, md_len - 1, "%s-%016lx", lprof->lp_md, cfg_instance);
1402 /* connections, registrations, sb setup */
1403 err = client_common_fill_super(sb, md, dt);
1405 GOTO(out_free_md, err);
1407 sbi->ll_client_common_fill_super_succeeded = 1;
1411 OBD_FREE(md, md_len);
1414 OBD_FREE(dt, dt_len);
1417 class_put_profile(lprof);
1420 ll_debugfs_unregister_super(sb);
1427 else if (test_bit(LL_SBI_VERBOSE, sbi->ll_flags))
1428 LCONSOLE_WARN("Mounted %s\n", profilenm);
1430 } /* ll_fill_super */
1432 void ll_put_super(struct super_block *sb)
1434 struct config_llog_instance cfg, params_cfg;
1435 struct obd_device *obd;
1436 struct lustre_sb_info *lsi = s2lsi(sb);
1437 struct ll_sb_info *sbi = ll_s2sbi(sb);
1438 char *profilenm = get_profile_name(sb);
1439 unsigned long cfg_instance = ll_get_cfg_instance(sb);
1441 int next, force = 1, rc = 0;
1445 GOTO(out_no_sbi, 0);
1447 /* Should replace instance_id with something better for ASLR */
1448 CDEBUG(D_VFSTRACE, "VFS Op: cfg_instance %s-%016lx (sb %p)\n",
1449 profilenm, cfg_instance, sb);
1451 cfg.cfg_instance = cfg_instance;
1452 lustre_end_log(sb, profilenm, &cfg);
1454 params_cfg.cfg_instance = cfg_instance;
1455 lustre_end_log(sb, PARAMS_FILENAME, ¶ms_cfg);
1457 if (sbi->ll_md_exp) {
1458 obd = class_exp2obd(sbi->ll_md_exp);
1460 force = obd->obd_force;
1463 /* Wait for unstable pages to be committed to stable storage */
1465 rc = l_wait_event_abortable(
1466 sbi->ll_cache->ccc_unstable_waitq,
1467 atomic_long_read(&sbi->ll_cache->ccc_unstable_nr) == 0);
1470 ccc_count = atomic_long_read(&sbi->ll_cache->ccc_unstable_nr);
1471 if (force == 0 && rc != -ERESTARTSYS)
1472 LASSERTF(ccc_count == 0, "count: %li\n", ccc_count);
1474 /* We need to set force before the lov_disconnect in
1475 * lustre_common_put_super, since l_d cleans up osc's as well.
1479 while ((obd = class_devices_in_group(&sbi->ll_sb_uuid,
1481 obd->obd_force = force;
1485 if (sbi->ll_client_common_fill_super_succeeded) {
1486 /* Only if client_common_fill_super succeeded */
1487 client_common_put_super(sb);
1490 /* imitate failed cleanup */
1491 if (OBD_FAIL_CHECK(OBD_FAIL_OBD_CLEANUP))
1495 while ((obd = class_devices_in_group(&sbi->ll_sb_uuid, &next)))
1496 class_manual_cleanup(obd);
1499 if (test_bit(LL_SBI_VERBOSE, sbi->ll_flags))
1500 LCONSOLE_WARN("Unmounted %s\n", profilenm ? profilenm : "");
1503 class_del_profile(profilenm);
1505 #ifndef HAVE_SUPER_SETUP_BDI_NAME
1506 if (lsi->lsi_flags & LSI_BDI_INITIALIZED) {
1507 bdi_destroy(&lsi->lsi_bdi);
1508 lsi->lsi_flags &= ~LSI_BDI_INITIALIZED;
1512 llcrypt_free_dummy_context(&lsi->lsi_dummy_enc_ctx);
1514 lsi->lsi_llsbi = NULL;
1516 lustre_common_put_super(sb);
1518 cl_env_cache_purge(~0);
1521 } /* client_put_super */
1523 struct inode *ll_inode_from_resource_lock(struct ldlm_lock *lock)
1525 struct inode *inode = NULL;
1527 /* NOTE: we depend on atomic igrab() -bzzz */
1528 lock_res_and_lock(lock);
1529 if (lock->l_resource->lr_lvb_inode) {
1530 struct ll_inode_info * lli;
1531 lli = ll_i2info(lock->l_resource->lr_lvb_inode);
1532 if (lli->lli_inode_magic == LLI_INODE_MAGIC) {
1533 inode = igrab(lock->l_resource->lr_lvb_inode);
1535 inode = lock->l_resource->lr_lvb_inode;
1536 LDLM_DEBUG_LIMIT(inode->i_state & I_FREEING ? D_INFO :
1537 D_WARNING, lock, "lr_lvb_inode %p is "
1538 "bogus: magic %08x",
1539 lock->l_resource->lr_lvb_inode,
1540 lli->lli_inode_magic);
1544 unlock_res_and_lock(lock);
1548 void ll_dir_clear_lsm_md(struct inode *inode)
1550 struct ll_inode_info *lli = ll_i2info(inode);
1552 LASSERT(S_ISDIR(inode->i_mode));
1554 if (lli->lli_lsm_md) {
1555 lmv_free_memmd(lli->lli_lsm_md);
1556 lli->lli_lsm_md = NULL;
1559 if (lli->lli_default_lsm_md) {
1560 lmv_free_memmd(lli->lli_default_lsm_md);
1561 lli->lli_default_lsm_md = NULL;
1565 static struct inode *ll_iget_anon_dir(struct super_block *sb,
1566 const struct lu_fid *fid,
1567 struct lustre_md *md)
1569 struct ll_sb_info *sbi = ll_s2sbi(sb);
1570 struct ll_inode_info *lli;
1571 struct mdt_body *body = md->body;
1572 struct inode *inode;
1578 ino = cl_fid_build_ino(fid, test_bit(LL_SBI_32BIT_API, sbi->ll_flags));
1579 inode = iget_locked(sb, ino);
1580 if (inode == NULL) {
1581 CERROR("%s: failed get simple inode "DFID": rc = -ENOENT\n",
1582 sbi->ll_fsname, PFID(fid));
1583 RETURN(ERR_PTR(-ENOENT));
1586 lli = ll_i2info(inode);
1587 if (inode->i_state & I_NEW) {
1588 inode->i_mode = (inode->i_mode & ~S_IFMT) |
1589 (body->mbo_mode & S_IFMT);
1590 LASSERTF(S_ISDIR(inode->i_mode), "Not slave inode "DFID"\n",
1593 inode->i_mtime.tv_sec = 0;
1594 inode->i_atime.tv_sec = 0;
1595 inode->i_ctime.tv_sec = 0;
1598 #ifdef HAVE_BACKING_DEV_INFO
1599 /* initializing backing dev info. */
1600 inode->i_mapping->backing_dev_info =
1601 &s2lsi(inode->i_sb)->lsi_bdi;
1603 inode->i_op = &ll_dir_inode_operations;
1604 inode->i_fop = &ll_dir_operations;
1605 lli->lli_fid = *fid;
1608 /* master object FID */
1609 lli->lli_pfid = body->mbo_fid1;
1610 CDEBUG(D_INODE, "lli %p slave "DFID" master "DFID"\n",
1611 lli, PFID(fid), PFID(&lli->lli_pfid));
1612 unlock_new_inode(inode);
1614 /* in directory restripe/auto-split, a directory will be
1615 * transformed to a stripe if it's plain, set its pfid here,
1616 * otherwise ll_lock_cancel_bits() can't find the master inode.
1618 lli->lli_pfid = body->mbo_fid1;
1624 static int ll_init_lsm_md(struct inode *inode, struct lustre_md *md)
1627 struct lmv_stripe_md *lsm = md->lmv;
1628 struct ll_inode_info *lli = ll_i2info(inode);
1631 LASSERT(lsm != NULL);
1633 CDEBUG(D_INODE, "%s: "DFID" set dir layout:\n",
1634 ll_i2sbi(inode)->ll_fsname, PFID(&lli->lli_fid));
1635 lsm_md_dump(D_INODE, lsm);
1637 if (!lmv_dir_striped(lsm))
1640 /* XXX sigh, this lsm_root initialization should be in
1641 * LMV layer, but it needs ll_iget right now, so we
1642 * put this here right now. */
1643 for (i = 0; i < lsm->lsm_md_stripe_count; i++) {
1644 fid = &lsm->lsm_md_oinfo[i].lmo_fid;
1645 LASSERT(lsm->lsm_md_oinfo[i].lmo_root == NULL);
1647 if (!fid_is_sane(fid))
1650 /* Unfortunately ll_iget will call ll_update_inode,
1651 * where the initialization of slave inode is slightly
1652 * different, so it reset lsm_md to NULL to avoid
1653 * initializing lsm for slave inode. */
1654 lsm->lsm_md_oinfo[i].lmo_root =
1655 ll_iget_anon_dir(inode->i_sb, fid, md);
1656 if (IS_ERR(lsm->lsm_md_oinfo[i].lmo_root)) {
1657 int rc = PTR_ERR(lsm->lsm_md_oinfo[i].lmo_root);
1659 lsm->lsm_md_oinfo[i].lmo_root = NULL;
1661 iput(lsm->lsm_md_oinfo[i].lmo_root);
1662 lsm->lsm_md_oinfo[i].lmo_root = NULL;
1668 lli->lli_lsm_md = lsm;
1673 static void ll_update_default_lsm_md(struct inode *inode, struct lustre_md *md)
1675 struct ll_inode_info *lli = ll_i2info(inode);
1679 if (!md->default_lmv) {
1680 /* clear default lsm */
1681 if (lli->lli_default_lsm_md) {
1682 down_write(&lli->lli_lsm_sem);
1683 if (lli->lli_default_lsm_md) {
1684 lmv_free_memmd(lli->lli_default_lsm_md);
1685 lli->lli_default_lsm_md = NULL;
1687 lli->lli_inherit_depth = 0;
1688 up_write(&lli->lli_lsm_sem);
1693 if (lli->lli_default_lsm_md) {
1694 /* do nonthing if default lsm isn't changed */
1695 down_read(&lli->lli_lsm_sem);
1696 if (lli->lli_default_lsm_md &&
1697 lsm_md_eq(lli->lli_default_lsm_md, md->default_lmv)) {
1698 up_read(&lli->lli_lsm_sem);
1701 up_read(&lli->lli_lsm_sem);
1704 down_write(&lli->lli_lsm_sem);
1705 if (lli->lli_default_lsm_md)
1706 lmv_free_memmd(lli->lli_default_lsm_md);
1707 lli->lli_default_lsm_md = md->default_lmv;
1708 lsm_md_dump(D_INODE, md->default_lmv);
1709 md->default_lmv = NULL;
1710 up_write(&lli->lli_lsm_sem);
1714 static int ll_update_lsm_md(struct inode *inode, struct lustre_md *md)
1716 struct ll_inode_info *lli = ll_i2info(inode);
1717 struct lmv_stripe_md *lsm = md->lmv;
1718 struct cl_attr *attr;
1723 LASSERT(S_ISDIR(inode->i_mode));
1724 CDEBUG(D_INODE, "update lsm %p of "DFID"\n", lli->lli_lsm_md,
1725 PFID(ll_inode2fid(inode)));
1727 /* update default LMV */
1728 if (md->default_lmv)
1729 ll_update_default_lsm_md(inode, md);
1731 /* after dir migration/restripe, a stripe may be turned into a
1732 * directory, in this case, zero out its lli_pfid.
1734 if (unlikely(fid_is_norm(&lli->lli_pfid)))
1735 fid_zero(&lli->lli_pfid);
1738 * no striped information from request, lustre_md from req does not
1739 * include stripeEA, see ll_md_setattr()
1745 * normally dir layout doesn't change, only take read lock to check
1746 * that to avoid blocking other MD operations.
1748 down_read(&lli->lli_lsm_sem);
1750 /* some current lookup initialized lsm, and unchanged */
1751 if (lli->lli_lsm_md && lsm_md_eq(lli->lli_lsm_md, lsm))
1752 GOTO(unlock, rc = 0);
1754 /* if dir layout doesn't match, check whether version is increased,
1755 * which means layout is changed, this happens in dir split/merge and
1758 * foreign LMV should not change.
1760 if (lli->lli_lsm_md && lmv_dir_striped(lli->lli_lsm_md) &&
1761 lsm->lsm_md_layout_version <=
1762 lli->lli_lsm_md->lsm_md_layout_version) {
1763 CERROR("%s: "DFID" dir layout mismatch:\n",
1764 ll_i2sbi(inode)->ll_fsname, PFID(&lli->lli_fid));
1765 lsm_md_dump(D_ERROR, lli->lli_lsm_md);
1766 lsm_md_dump(D_ERROR, lsm);
1767 GOTO(unlock, rc = -EINVAL);
1770 up_read(&lli->lli_lsm_sem);
1771 down_write(&lli->lli_lsm_sem);
1772 /* clear existing lsm */
1773 if (lli->lli_lsm_md) {
1774 lmv_free_memmd(lli->lli_lsm_md);
1775 lli->lli_lsm_md = NULL;
1778 rc = ll_init_lsm_md(inode, md);
1780 up_write(&lli->lli_lsm_sem);
1784 /* md_merge_attr() may take long, since lsm is already set, switch to
1787 downgrade_write(&lli->lli_lsm_sem);
1789 /* set md->lmv to NULL, so the following free lustre_md will not free
1794 if (!lmv_dir_striped(lli->lli_lsm_md))
1795 GOTO(unlock, rc = 0);
1797 OBD_ALLOC_PTR(attr);
1799 GOTO(unlock, rc = -ENOMEM);
1801 /* validate the lsm */
1802 rc = md_merge_attr(ll_i2mdexp(inode), lli->lli_lsm_md, attr,
1803 ll_md_blocking_ast);
1805 if (md->body->mbo_valid & OBD_MD_FLNLINK)
1806 md->body->mbo_nlink = attr->cat_nlink;
1807 if (md->body->mbo_valid & OBD_MD_FLSIZE)
1808 md->body->mbo_size = attr->cat_size;
1809 if (md->body->mbo_valid & OBD_MD_FLATIME)
1810 md->body->mbo_atime = attr->cat_atime;
1811 if (md->body->mbo_valid & OBD_MD_FLCTIME)
1812 md->body->mbo_ctime = attr->cat_ctime;
1813 if (md->body->mbo_valid & OBD_MD_FLMTIME)
1814 md->body->mbo_mtime = attr->cat_mtime;
1820 up_read(&lli->lli_lsm_sem);
1825 void ll_clear_inode(struct inode *inode)
1827 struct ll_inode_info *lli = ll_i2info(inode);
1828 struct ll_sb_info *sbi = ll_i2sbi(inode);
1832 CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p)\n",
1833 PFID(ll_inode2fid(inode)), inode);
1835 if (S_ISDIR(inode->i_mode)) {
1836 /* these should have been cleared in ll_file_release */
1837 LASSERT(lli->lli_opendir_key == NULL);
1838 LASSERT(lli->lli_sai == NULL);
1839 LASSERT(lli->lli_opendir_pid == 0);
1841 pcc_inode_free(inode);
1844 md_null_inode(sbi->ll_md_exp, ll_inode2fid(inode));
1846 LASSERT(!lli->lli_open_fd_write_count);
1847 LASSERT(!lli->lli_open_fd_read_count);
1848 LASSERT(!lli->lli_open_fd_exec_count);
1850 if (lli->lli_mds_write_och)
1851 ll_md_real_close(inode, FMODE_WRITE);
1852 if (lli->lli_mds_exec_och)
1853 ll_md_real_close(inode, FMODE_EXEC);
1854 if (lli->lli_mds_read_och)
1855 ll_md_real_close(inode, FMODE_READ);
1857 if (S_ISLNK(inode->i_mode) && lli->lli_symlink_name) {
1858 OBD_FREE(lli->lli_symlink_name,
1859 strlen(lli->lli_symlink_name) + 1);
1860 lli->lli_symlink_name = NULL;
1863 ll_xattr_cache_destroy(inode);
1865 forget_all_cached_acls(inode);
1867 lli->lli_inode_magic = LLI_INODE_DEAD;
1869 if (S_ISDIR(inode->i_mode))
1870 ll_dir_clear_lsm_md(inode);
1871 else if (S_ISREG(inode->i_mode) && !is_bad_inode(inode))
1872 LASSERT(list_empty(&lli->lli_agl_list));
1875 * XXX This has to be done before lsm is freed below, because
1876 * cl_object still uses inode lsm.
1878 cl_inode_fini(inode);
1880 llcrypt_put_encryption_info(inode);
1885 static int ll_md_setattr(struct dentry *dentry, struct md_op_data *op_data)
1887 struct lustre_md md;
1888 struct inode *inode = dentry->d_inode;
1889 struct ll_sb_info *sbi = ll_i2sbi(inode);
1890 struct ptlrpc_request *request = NULL;
1895 op_data = ll_prep_md_op_data(op_data, inode, NULL, NULL, 0, 0,
1896 LUSTRE_OPC_ANY, NULL);
1897 if (IS_ERR(op_data))
1898 RETURN(PTR_ERR(op_data));
1900 /* If this is a chgrp of a regular file, we want to reserve enough
1901 * quota to cover the entire file size.
1903 if (S_ISREG(inode->i_mode) && op_data->op_attr.ia_valid & ATTR_GID &&
1904 from_kgid(&init_user_ns, op_data->op_attr.ia_gid) !=
1905 from_kgid(&init_user_ns, inode->i_gid)) {
1906 op_data->op_xvalid |= OP_XVALID_BLOCKS;
1907 op_data->op_attr_blocks = inode->i_blocks;
1911 rc = md_setattr(sbi->ll_md_exp, op_data, NULL, 0, &request);
1913 ptlrpc_req_finished(request);
1914 if (rc == -ENOENT) {
1916 /* Unlinked special device node? Or just a race?
1917 * Pretend we done everything. */
1918 if (!S_ISREG(inode->i_mode) &&
1919 !S_ISDIR(inode->i_mode)) {
1920 ia_valid = op_data->op_attr.ia_valid;
1921 op_data->op_attr.ia_valid &= ~TIMES_SET_FLAGS;
1922 rc = simple_setattr(&init_user_ns, dentry,
1924 op_data->op_attr.ia_valid = ia_valid;
1926 } else if (rc != -EPERM && rc != -EACCES && rc != -ETXTBSY) {
1927 CERROR("md_setattr fails: rc = %d\n", rc);
1932 rc = md_get_lustre_md(sbi->ll_md_exp, &request->rq_pill, sbi->ll_dt_exp,
1933 sbi->ll_md_exp, &md);
1935 ptlrpc_req_finished(request);
1939 ia_valid = op_data->op_attr.ia_valid;
1940 /* inode size will be in ll_setattr_ost, can't do it now since dirty
1941 * cache is not cleared yet. */
1942 op_data->op_attr.ia_valid &= ~(TIMES_SET_FLAGS | ATTR_SIZE);
1943 if (S_ISREG(inode->i_mode))
1945 rc = simple_setattr(&init_user_ns, dentry, &op_data->op_attr);
1946 if (S_ISREG(inode->i_mode))
1947 inode_unlock(inode);
1948 op_data->op_attr.ia_valid = ia_valid;
1950 rc = ll_update_inode(inode, &md);
1951 ptlrpc_req_finished(request);
1957 * Zero portion of page that is part of @inode.
1958 * This implies, if necessary:
1959 * - taking cl_lock on range corresponding to concerned page
1960 * - grabbing vm page
1961 * - associating cl_page
1962 * - proceeding to clio read
1963 * - zeroing range in page
1964 * - proceeding to cl_page flush
1965 * - releasing cl_lock
1967 * \param[in] inode inode
1968 * \param[in] index page index
1969 * \param[in] offset offset in page to start zero from
1970 * \param[in] len len to zero
1972 * \retval 0 on success
1973 * \retval negative errno on failure
1975 int ll_io_zero_page(struct inode *inode, pgoff_t index, pgoff_t offset,
1978 struct ll_inode_info *lli = ll_i2info(inode);
1979 struct cl_object *clob = lli->lli_clob;
1981 struct lu_env *env = NULL;
1982 struct cl_io *io = NULL;
1983 struct cl_page *clpage = NULL;
1984 struct page *vmpage = NULL;
1985 unsigned from = index << PAGE_SHIFT;
1986 struct cl_lock *lock = NULL;
1987 struct cl_lock_descr *descr = NULL;
1988 struct cl_2queue *queue = NULL;
1989 struct cl_sync_io *anchor = NULL;
1990 bool holdinglock = false;
1995 env = cl_env_get(&refcheck);
1997 RETURN(PTR_ERR(env));
1999 io = vvp_env_thread_io(env);
2001 rc = cl_io_rw_init(env, io, CIT_WRITE, from, PAGE_SIZE);
2005 lock = vvp_env_lock(env);
2006 descr = &lock->cll_descr;
2007 descr->cld_obj = io->ci_obj;
2008 descr->cld_start = cl_index(io->ci_obj, from);
2009 descr->cld_end = cl_index(io->ci_obj, from + PAGE_SIZE - 1);
2010 descr->cld_mode = CLM_WRITE;
2011 descr->cld_enq_flags = CEF_MUST | CEF_NONBLOCK;
2013 /* request lock for page */
2014 rc = cl_lock_request(env, io, lock);
2015 /* -ECANCELED indicates a matching lock with a different extent
2016 * was already present, and -EEXIST indicates a matching lock
2017 * on exactly the same extent was already present.
2018 * In both cases it means we are covered.
2020 if (rc == -ECANCELED || rc == -EEXIST)
2028 vmpage = grab_cache_page_nowait(inode->i_mapping, index);
2030 GOTO(rellock, rc = -EOPNOTSUPP);
2032 if (!PageDirty(vmpage)) {
2033 /* associate cl_page */
2034 clpage = cl_page_find(env, clob, vmpage->index,
2035 vmpage, CPT_CACHEABLE);
2037 GOTO(pagefini, rc = PTR_ERR(clpage));
2039 cl_page_assume(env, io, clpage);
2042 if (!PageUptodate(vmpage) && !PageDirty(vmpage) &&
2043 !PageWriteback(vmpage)) {
2045 /* Set PagePrivate2 to detect special case of empty page
2046 * in osc_brw_fini_request().
2047 * It is also used to tell ll_io_read_page() that we do not
2048 * want the vmpage to be unlocked.
2050 SetPagePrivate2(vmpage);
2051 rc = ll_io_read_page(env, io, clpage, NULL);
2052 if (!PagePrivate2(vmpage)) {
2053 /* PagePrivate2 was cleared in osc_brw_fini_request()
2054 * meaning we read an empty page. In this case, in order
2055 * to avoid allocating unnecessary block in truncated
2056 * file, we must not zero and write as below. Subsequent
2057 * server-side truncate will handle things correctly.
2059 cl_page_unassume(env, io, clpage);
2060 GOTO(clpfini, rc = 0);
2062 ClearPagePrivate2(vmpage);
2067 /* Thanks to PagePrivate2 flag, ll_io_read_page() did not unlock
2068 * the vmpage, so we are good to proceed and zero range in page.
2070 zero_user(vmpage, offset, len);
2072 if (holdinglock && clpage) {
2073 /* explicitly write newly modified page */
2074 queue = &io->ci_queue;
2075 cl_2queue_init(queue);
2076 anchor = &vvp_env_info(env)->vti_anchor;
2077 cl_sync_io_init(anchor, 1);
2078 clpage->cp_sync_io = anchor;
2079 cl_page_list_add(&queue->c2_qin, clpage, true);
2080 rc = cl_io_submit_rw(env, io, CRT_WRITE, queue);
2082 GOTO(queuefini1, rc);
2083 rc = cl_sync_io_wait(env, anchor, 0);
2085 GOTO(queuefini2, rc);
2086 cl_page_assume(env, io, clpage);
2089 cl_2queue_discard(env, io, queue);
2091 cl_2queue_disown(env, queue);
2092 cl_2queue_fini(env, queue);
2097 cl_page_put(env, clpage);
2099 unlock_page(vmpage);
2103 cl_lock_release(env, lock);
2105 cl_io_fini(env, io);
2108 cl_env_put(env, &refcheck);
2114 * Get reference file from volatile file name.
2115 * Volatile file name may look like:
2116 * <parent>/LUSTRE_VOLATILE_HDR:<mdt_index>:<random>:fd=<fd>
2117 * where fd is opened descriptor of reference file.
2119 * \param[in] volatile_name volatile file name
2120 * \param[in] volatile_len volatile file name length
2121 * \param[out] ref_file pointer to struct file of reference file
2123 * \retval 0 on success
2124 * \retval negative errno on failure
2126 int volatile_ref_file(const char *volatile_name, int volatile_len,
2127 struct file **ref_file)
2129 char *p, *q, *fd_str;
2132 p = strnstr(volatile_name, ":fd=", volatile_len);
2133 if (!p || strlen(p + 4) == 0)
2136 q = strchrnul(p + 4, ':');
2137 fd_str = kstrndup(p + 4, q - p - 4, GFP_NOFS);
2140 rc = kstrtouint(fd_str, 10, &fd);
2145 *ref_file = fget(fd);
2151 /* If this inode has objects allocated to it (lsm != NULL), then the OST
2152 * object(s) determine the file size and mtime. Otherwise, the MDS will
2153 * keep these values until such a time that objects are allocated for it.
2154 * We do the MDS operations first, as it is checking permissions for us.
2155 * We don't to the MDS RPC if there is nothing that we want to store there,
2156 * otherwise there is no harm in updating mtime/atime on the MDS if we are
2157 * going to do an RPC anyways.
2159 * If we are doing a truncate, we will send the mtime and ctime updates
2160 * to the OST with the punch RPC, otherwise we do an explicit setattr RPC.
2161 * I don't believe it is possible to get e.g. ATTR_MTIME_SET and ATTR_SIZE
2164 * In case of HSMimport, we only set attr on MDS.
2166 int ll_setattr_raw(struct dentry *dentry, struct iattr *attr,
2167 enum op_xvalid xvalid, bool hsm_import)
2169 struct inode *inode = dentry->d_inode;
2170 struct ll_inode_info *lli = ll_i2info(inode);
2171 struct md_op_data *op_data = NULL;
2172 ktime_t kstart = ktime_get();
2177 CDEBUG(D_VFSTRACE, "%s: setattr inode "DFID"(%p) from %llu to %llu, "
2178 "valid %x, hsm_import %d\n",
2179 ll_i2sbi(inode)->ll_fsname, PFID(&lli->lli_fid),
2180 inode, i_size_read(inode), attr->ia_size, attr->ia_valid,
2183 if (attr->ia_valid & ATTR_SIZE) {
2184 /* Check new size against VFS/VM file size limit and rlimit */
2185 rc = inode_newsize_ok(inode, attr->ia_size);
2189 /* The maximum Lustre file size is variable, based on the
2190 * OST maximum object size and number of stripes. This
2191 * needs another check in addition to the VFS check above. */
2192 if (attr->ia_size > ll_file_maxbytes(inode)) {
2193 CDEBUG(D_INODE,"file "DFID" too large %llu > %llu\n",
2194 PFID(&lli->lli_fid), attr->ia_size,
2195 ll_file_maxbytes(inode));
2199 attr->ia_valid |= ATTR_MTIME | ATTR_CTIME;
2202 /* POSIX: check before ATTR_*TIME_SET set (from inode_change_ok) */
2203 if (attr->ia_valid & TIMES_SET_FLAGS) {
2204 if ((!uid_eq(current_fsuid(), inode->i_uid)) &&
2205 !capable(CAP_FOWNER))
2209 /* We mark all of the fields "set" so MDS/OST does not re-set them */
2210 if (!(xvalid & OP_XVALID_CTIME_SET) &&
2211 (attr->ia_valid & ATTR_CTIME)) {
2212 attr->ia_ctime = current_time(inode);
2213 xvalid |= OP_XVALID_CTIME_SET;
2215 if (!(attr->ia_valid & ATTR_ATIME_SET) &&
2216 (attr->ia_valid & ATTR_ATIME)) {
2217 attr->ia_atime = current_time(inode);
2218 attr->ia_valid |= ATTR_ATIME_SET;
2220 if (!(attr->ia_valid & ATTR_MTIME_SET) &&
2221 (attr->ia_valid & ATTR_MTIME)) {
2222 attr->ia_mtime = current_time(inode);
2223 attr->ia_valid |= ATTR_MTIME_SET;
2226 if (attr->ia_valid & (ATTR_MTIME | ATTR_CTIME))
2227 CDEBUG(D_INODE, "setting mtime %lld, ctime %lld, now = %lld\n",
2228 (s64)attr->ia_mtime.tv_sec, (s64)attr->ia_ctime.tv_sec,
2229 ktime_get_real_seconds());
2231 if (S_ISREG(inode->i_mode))
2232 inode_unlock(inode);
2234 /* We always do an MDS RPC, even if we're only changing the size;
2235 * only the MDS knows whether truncate() should fail with -ETXTBUSY */
2237 OBD_ALLOC_PTR(op_data);
2238 if (op_data == NULL)
2239 GOTO(out, rc = -ENOMEM);
2241 if (!hsm_import && attr->ia_valid & ATTR_SIZE) {
2242 /* If we are changing file size, file content is
2243 * modified, flag it.
2245 xvalid |= OP_XVALID_OWNEROVERRIDE;
2246 op_data->op_bias |= MDS_DATA_MODIFIED;
2247 clear_bit(LLIF_DATA_MODIFIED, &lli->lli_flags);
2250 if (attr->ia_valid & ATTR_FILE) {
2251 struct ll_file_data *fd = attr->ia_file->private_data;
2253 if (fd->fd_lease_och)
2254 op_data->op_bias |= MDS_TRUNC_KEEP_LEASE;
2257 op_data->op_attr = *attr;
2258 op_data->op_xvalid = xvalid;
2260 rc = ll_md_setattr(dentry, op_data);
2264 if (!S_ISREG(inode->i_mode) || hsm_import)
2267 if (attr->ia_valid & (ATTR_SIZE | ATTR_ATIME | ATTR_ATIME_SET |
2268 ATTR_MTIME | ATTR_MTIME_SET | ATTR_CTIME) ||
2269 xvalid & OP_XVALID_CTIME_SET) {
2270 bool cached = false;
2272 rc = pcc_inode_setattr(inode, attr, &cached);
2275 CERROR("%s: PCC inode "DFID" setattr failed: "
2277 ll_i2sbi(inode)->ll_fsname,
2278 PFID(&lli->lli_fid), rc);
2282 unsigned int flags = 0;
2284 /* For truncate and utimes sending attributes to OSTs,
2285 * setting mtime/atime to the past will be performed
2286 * under PW [0:EOF] extent lock (new_size:EOF for
2287 * truncate). It may seem excessive to send mtime/atime
2288 * updates to OSTs when not setting times to past, but
2289 * it is necessary due to possible time
2290 * de-synchronization between MDT inode and OST objects
2292 if (S_ISREG(inode->i_mode) && IS_ENCRYPTED(inode)) {
2293 xvalid |= OP_XVALID_FLAGS;
2294 flags = LUSTRE_ENCRYPT_FL;
2295 /* Call to ll_io_zero_page is not necessary if
2296 * truncating on PAGE_SIZE boundary, because
2297 * whole pages will be wiped.
2298 * In case of Direct IO, all we need is to set
2301 if (attr->ia_valid & ATTR_SIZE &&
2302 attr->ia_size & ~PAGE_MASK &&
2303 !(attr->ia_valid & ATTR_FILE &&
2304 attr->ia_file->f_flags & O_DIRECT)) {
2306 attr->ia_size & (PAGE_SIZE - 1);
2308 rc = ll_io_zero_page(inode,
2309 attr->ia_size >> PAGE_SHIFT,
2310 offset, PAGE_SIZE - offset);
2314 /* If encrypted volatile file without the key,
2315 * we need to fetch size from reference file,
2316 * and set it on OST objects. This happens when
2317 * migrating or extending an encrypted file
2320 if (filename_is_volatile(dentry->d_name.name,
2323 llcrypt_require_key(inode) == -ENOKEY) {
2324 struct file *ref_file;
2325 struct inode *ref_inode;
2326 struct ll_inode_info *ref_lli;
2327 struct cl_object *ref_obj;
2328 struct cl_attr ref_attr = { 0 };
2332 rc = volatile_ref_file(
2333 dentry->d_name.name,
2339 ref_inode = file_inode(ref_file);
2342 GOTO(out, rc = -EINVAL);
2345 env = cl_env_get(&refcheck);
2347 GOTO(out, rc = PTR_ERR(env));
2349 ref_lli = ll_i2info(ref_inode);
2350 ref_obj = ref_lli->lli_clob;
2351 cl_object_attr_lock(ref_obj);
2352 rc = cl_object_attr_get(env, ref_obj,
2354 cl_object_attr_unlock(ref_obj);
2355 cl_env_put(env, &refcheck);
2360 attr->ia_valid |= ATTR_SIZE;
2361 attr->ia_size = ref_attr.cat_size;
2364 rc = cl_setattr_ost(lli->lli_clob, attr, xvalid, flags);
2368 /* If the file was restored, it needs to set dirty flag.
2370 * We've already sent MDS_DATA_MODIFIED flag in
2371 * ll_md_setattr() for truncate. However, the MDT refuses to
2372 * set the HS_DIRTY flag on released files, so we have to set
2373 * it again if the file has been restored. Please check how
2374 * LLIF_DATA_MODIFIED is set in vvp_io_setattr_fini().
2376 * Please notice that if the file is not released, the previous
2377 * MDS_DATA_MODIFIED has taken effect and usually
2378 * LLIF_DATA_MODIFIED is not set(see vvp_io_setattr_fini()).
2379 * This way we can save an RPC for common open + trunc
2381 if (test_and_clear_bit(LLIF_DATA_MODIFIED, &lli->lli_flags)) {
2382 struct hsm_state_set hss = {
2383 .hss_valid = HSS_SETMASK,
2384 .hss_setmask = HS_DIRTY,
2388 rc2 = ll_hsm_state_set(inode, &hss);
2389 /* truncate and write can happen at the same time, so that
2390 * the file can be set modified even though the file is not
2391 * restored from released state, and ll_hsm_state_set() is
2392 * not applicable for the file, and rc2 < 0 is normal in this
2395 CDEBUG(D_INFO, DFID "HSM set dirty failed: rc2 = %d\n",
2396 PFID(ll_inode2fid(inode)), rc2);
2401 if (op_data != NULL)
2402 ll_finish_md_op_data(op_data);
2404 if (S_ISREG(inode->i_mode)) {
2406 if ((attr->ia_valid & ATTR_SIZE) && !hsm_import)
2407 inode_dio_wait(inode);
2408 /* Once we've got the i_mutex, it's safe to set the S_NOSEC
2409 * flag. ll_update_inode (called from ll_md_setattr), clears
2410 * inode flags, so there is a gap where S_NOSEC is not set.
2411 * This can cause a writer to take the i_mutex unnecessarily,
2412 * but this is safe to do and should be rare. */
2413 inode_has_no_xattr(inode);
2417 ll_stats_ops_tally(ll_i2sbi(inode), attr->ia_valid & ATTR_SIZE ?
2418 LPROC_LL_TRUNC : LPROC_LL_SETATTR,
2419 ktime_us_delta(ktime_get(), kstart));
2424 int ll_setattr(struct user_namespace *mnt_userns, struct dentry *de,
2427 int mode = de->d_inode->i_mode;
2428 enum op_xvalid xvalid = 0;
2431 rc = llcrypt_prepare_setattr(de, attr);
2435 if ((attr->ia_valid & (ATTR_CTIME|ATTR_SIZE|ATTR_MODE)) ==
2436 (ATTR_CTIME|ATTR_SIZE|ATTR_MODE))
2437 xvalid |= OP_XVALID_OWNEROVERRIDE;
2439 if (((attr->ia_valid & (ATTR_MODE|ATTR_FORCE|ATTR_SIZE)) ==
2440 (ATTR_SIZE|ATTR_MODE)) &&
2441 (((mode & S_ISUID) && !(attr->ia_mode & S_ISUID)) ||
2442 (((mode & (S_ISGID|S_IXGRP)) == (S_ISGID|S_IXGRP)) &&
2443 !(attr->ia_mode & S_ISGID))))
2444 attr->ia_valid |= ATTR_FORCE;
2446 if ((attr->ia_valid & ATTR_MODE) &&
2448 !(attr->ia_mode & S_ISUID) &&
2449 !(attr->ia_valid & ATTR_KILL_SUID))
2450 attr->ia_valid |= ATTR_KILL_SUID;
2452 if ((attr->ia_valid & ATTR_MODE) &&
2453 ((mode & (S_ISGID|S_IXGRP)) == (S_ISGID|S_IXGRP)) &&
2454 !(attr->ia_mode & S_ISGID) &&
2455 !(attr->ia_valid & ATTR_KILL_SGID))
2456 attr->ia_valid |= ATTR_KILL_SGID;
2458 return ll_setattr_raw(de, attr, xvalid, false);
2461 int ll_statfs_internal(struct ll_sb_info *sbi, struct obd_statfs *osfs,
2464 struct obd_statfs obd_osfs = { 0 };
2469 max_age = ktime_get_seconds() - sbi->ll_statfs_max_age;
2471 if (test_bit(LL_SBI_LAZYSTATFS, sbi->ll_flags))
2472 flags |= OBD_STATFS_NODELAY;
2474 rc = obd_statfs(NULL, sbi->ll_md_exp, osfs, max_age, flags);
2478 osfs->os_type = LL_SUPER_MAGIC;
2480 CDEBUG(D_SUPER, "MDC blocks %llu/%llu objects %llu/%llu\n",
2481 osfs->os_bavail, osfs->os_blocks, osfs->os_ffree, osfs->os_files);
2483 if (osfs->os_state & OS_STATFS_SUM)
2486 rc = obd_statfs(NULL, sbi->ll_dt_exp, &obd_osfs, max_age, flags);
2487 if (rc) /* Possibly a filesystem with no OSTs. Report MDT totals. */
2490 CDEBUG(D_SUPER, "OSC blocks %llu/%llu objects %llu/%llu\n",
2491 obd_osfs.os_bavail, obd_osfs.os_blocks, obd_osfs.os_ffree,
2494 osfs->os_bsize = obd_osfs.os_bsize;
2495 osfs->os_blocks = obd_osfs.os_blocks;
2496 osfs->os_bfree = obd_osfs.os_bfree;
2497 osfs->os_bavail = obd_osfs.os_bavail;
2499 /* If we have _some_ OSTs, but don't have as many free objects on the
2500 * OSTs as inodes on the MDTs, reduce the reported number of inodes
2501 * to compensate, so that the "inodes in use" number is correct.
2502 * This should be kept in sync with lod_statfs() behaviour.
2504 if (obd_osfs.os_files && obd_osfs.os_ffree < osfs->os_ffree) {
2505 osfs->os_files = (osfs->os_files - osfs->os_ffree) +
2507 osfs->os_ffree = obd_osfs.os_ffree;
2514 static int ll_statfs_project(struct inode *inode, struct kstatfs *sfs)
2516 struct if_quotactl qctl = {
2517 .qc_cmd = LUSTRE_Q_GETQUOTA,
2518 .qc_type = PRJQUOTA,
2519 .qc_valid = QC_GENERAL,
2521 u64 limit, curblock;
2524 qctl.qc_id = ll_i2info(inode)->lli_projid;
2525 ret = quotactl_ioctl(inode->i_sb, &qctl);
2527 /* ignore errors if project ID does not have
2528 * a quota limit or feature unsupported.
2530 if (ret == -ESRCH || ret == -EOPNOTSUPP)
2535 limit = ((qctl.qc_dqblk.dqb_bsoftlimit ?
2536 qctl.qc_dqblk.dqb_bsoftlimit :
2537 qctl.qc_dqblk.dqb_bhardlimit) * 1024) / sfs->f_bsize;
2538 if (limit && sfs->f_blocks > limit) {
2539 curblock = (qctl.qc_dqblk.dqb_curspace +
2540 sfs->f_bsize - 1) / sfs->f_bsize;
2541 sfs->f_blocks = limit;
2542 sfs->f_bfree = sfs->f_bavail =
2543 (sfs->f_blocks > curblock) ?
2544 (sfs->f_blocks - curblock) : 0;
2547 limit = qctl.qc_dqblk.dqb_isoftlimit ?
2548 qctl.qc_dqblk.dqb_isoftlimit :
2549 qctl.qc_dqblk.dqb_ihardlimit;
2550 if (limit && sfs->f_files > limit) {
2551 sfs->f_files = limit;
2552 sfs->f_ffree = (sfs->f_files >
2553 qctl.qc_dqblk.dqb_curinodes) ?
2554 (sfs->f_files - qctl.qc_dqblk.dqb_curinodes) : 0;
2560 int ll_statfs(struct dentry *de, struct kstatfs *sfs)
2562 struct super_block *sb = de->d_sb;
2563 struct obd_statfs osfs;
2564 __u64 fsid = huge_encode_dev(sb->s_dev);
2565 ktime_t kstart = ktime_get();
2568 CDEBUG(D_VFSTRACE, "VFS Op:sb=%s (%p)\n", sb->s_id, sb);
2570 /* Some amount of caching on the client is allowed */
2571 rc = ll_statfs_internal(ll_s2sbi(sb), &osfs, OBD_STATFS_SUM);
2575 statfs_unpack(sfs, &osfs);
2577 /* We need to downshift for all 32-bit kernels, because we can't
2578 * tell if the kernel is being called via sys_statfs64() or not.
2579 * Stop before overflowing f_bsize - in which case it is better
2580 * to just risk EOVERFLOW if caller is using old sys_statfs(). */
2581 if (sizeof(long) < 8) {
2582 while (osfs.os_blocks > ~0UL && sfs->f_bsize < 0x40000000) {
2585 osfs.os_blocks >>= 1;
2586 osfs.os_bfree >>= 1;
2587 osfs.os_bavail >>= 1;
2591 sfs->f_blocks = osfs.os_blocks;
2592 sfs->f_bfree = osfs.os_bfree;
2593 sfs->f_bavail = osfs.os_bavail;
2594 sfs->f_fsid.val[0] = (__u32)fsid;
2595 sfs->f_fsid.val[1] = (__u32)(fsid >> 32);
2596 if (ll_i2info(de->d_inode)->lli_projid)
2597 return ll_statfs_project(de->d_inode, sfs);
2599 ll_stats_ops_tally(ll_s2sbi(sb), LPROC_LL_STATFS,
2600 ktime_us_delta(ktime_get(), kstart));
2605 void ll_inode_size_lock(struct inode *inode)
2607 struct ll_inode_info *lli;
2609 LASSERT(!S_ISDIR(inode->i_mode));
2611 lli = ll_i2info(inode);
2612 mutex_lock(&lli->lli_size_mutex);
2615 void ll_inode_size_unlock(struct inode *inode)
2617 struct ll_inode_info *lli;
2619 lli = ll_i2info(inode);
2620 mutex_unlock(&lli->lli_size_mutex);
2623 void ll_update_inode_flags(struct inode *inode, unsigned int ext_flags)
2625 /* do not clear encryption flag */
2626 ext_flags |= ll_inode_to_ext_flags(inode->i_flags) & LUSTRE_ENCRYPT_FL;
2627 inode->i_flags = ll_ext_to_inode_flags(ext_flags);
2628 if (ext_flags & LUSTRE_PROJINHERIT_FL)
2629 set_bit(LLIF_PROJECT_INHERIT, &ll_i2info(inode)->lli_flags);
2631 clear_bit(LLIF_PROJECT_INHERIT, &ll_i2info(inode)->lli_flags);
2634 int ll_update_inode(struct inode *inode, struct lustre_md *md)
2636 struct ll_inode_info *lli = ll_i2info(inode);
2637 struct mdt_body *body = md->body;
2638 struct ll_sb_info *sbi = ll_i2sbi(inode);
2642 if (body->mbo_valid & OBD_MD_FLEASIZE) {
2643 rc = cl_file_inode_init(inode, md);
2648 if (S_ISDIR(inode->i_mode)) {
2649 rc = ll_update_lsm_md(inode, md);
2654 if (body->mbo_valid & OBD_MD_FLACL)
2655 lli_replace_acl(lli, md);
2657 api32 = test_bit(LL_SBI_32BIT_API, sbi->ll_flags);
2658 inode->i_ino = cl_fid_build_ino(&body->mbo_fid1, api32);
2659 inode->i_generation = cl_fid_build_gen(&body->mbo_fid1);
2661 if (body->mbo_valid & OBD_MD_FLATIME) {
2662 if (body->mbo_atime > inode->i_atime.tv_sec)
2663 inode->i_atime.tv_sec = body->mbo_atime;
2664 lli->lli_atime = body->mbo_atime;
2667 if (body->mbo_valid & OBD_MD_FLMTIME) {
2668 if (body->mbo_mtime > inode->i_mtime.tv_sec) {
2670 "setting ino %lu mtime from %lld to %llu\n",
2671 inode->i_ino, (s64)inode->i_mtime.tv_sec,
2673 inode->i_mtime.tv_sec = body->mbo_mtime;
2675 lli->lli_mtime = body->mbo_mtime;
2678 if (body->mbo_valid & OBD_MD_FLCTIME) {
2679 if (body->mbo_ctime > inode->i_ctime.tv_sec)
2680 inode->i_ctime.tv_sec = body->mbo_ctime;
2681 lli->lli_ctime = body->mbo_ctime;
2684 if (body->mbo_valid & OBD_MD_FLBTIME)
2685 lli->lli_btime = body->mbo_btime;
2687 /* Clear i_flags to remove S_NOSEC before permissions are updated */
2688 if (body->mbo_valid & OBD_MD_FLFLAGS)
2689 ll_update_inode_flags(inode, body->mbo_flags);
2690 if (body->mbo_valid & OBD_MD_FLMODE)
2691 inode->i_mode = (inode->i_mode & S_IFMT) |
2692 (body->mbo_mode & ~S_IFMT);
2694 if (body->mbo_valid & OBD_MD_FLTYPE)
2695 inode->i_mode = (inode->i_mode & ~S_IFMT) |
2696 (body->mbo_mode & S_IFMT);
2698 LASSERT(inode->i_mode != 0);
2699 if (body->mbo_valid & OBD_MD_FLUID)
2700 inode->i_uid = make_kuid(&init_user_ns, body->mbo_uid);
2701 if (body->mbo_valid & OBD_MD_FLGID)
2702 inode->i_gid = make_kgid(&init_user_ns, body->mbo_gid);
2703 if (body->mbo_valid & OBD_MD_FLPROJID)
2704 lli->lli_projid = body->mbo_projid;
2705 if (body->mbo_valid & OBD_MD_FLNLINK) {
2706 spin_lock(&inode->i_lock);
2707 set_nlink(inode, body->mbo_nlink);
2708 spin_unlock(&inode->i_lock);
2710 if (body->mbo_valid & OBD_MD_FLRDEV)
2711 inode->i_rdev = old_decode_dev(body->mbo_rdev);
2713 if (body->mbo_valid & OBD_MD_FLID) {
2714 /* FID shouldn't be changed! */
2715 if (fid_is_sane(&lli->lli_fid)) {
2716 LASSERTF(lu_fid_eq(&lli->lli_fid, &body->mbo_fid1),
2717 "Trying to change FID "DFID
2718 " to the "DFID", inode "DFID"(%p)\n",
2719 PFID(&lli->lli_fid), PFID(&body->mbo_fid1),
2720 PFID(ll_inode2fid(inode)), inode);
2722 lli->lli_fid = body->mbo_fid1;
2726 LASSERT(fid_seq(&lli->lli_fid) != 0);
2728 /* In case of encrypted file without the key, please do not lose
2729 * clear text size stored into lli_lazysize in ll_merge_attr(),
2730 * we will need it in ll_prepare_close().
2732 if (lli->lli_attr_valid & OBD_MD_FLLAZYSIZE && lli->lli_lazysize &&
2733 llcrypt_require_key(inode) == -ENOKEY)
2734 lli->lli_attr_valid = body->mbo_valid | OBD_MD_FLLAZYSIZE;
2736 lli->lli_attr_valid = body->mbo_valid;
2737 if (body->mbo_valid & OBD_MD_FLSIZE) {
2738 i_size_write(inode, body->mbo_size);
2740 CDEBUG(D_VFSTRACE, "inode="DFID", updating i_size %llu\n",
2741 PFID(ll_inode2fid(inode)),
2742 (unsigned long long)body->mbo_size);
2744 if (body->mbo_valid & OBD_MD_FLBLOCKS)
2745 inode->i_blocks = body->mbo_blocks;
2747 if (body->mbo_valid & OBD_MD_FLLAZYSIZE)
2748 lli->lli_lazysize = body->mbo_size;
2749 if (body->mbo_valid & OBD_MD_FLLAZYBLOCKS)
2750 lli->lli_lazyblocks = body->mbo_blocks;
2753 if (body->mbo_valid & OBD_MD_TSTATE) {
2754 /* Set LLIF_FILE_RESTORING if restore ongoing and
2755 * clear it when done to ensure to start again
2756 * glimpsing updated attrs
2758 if (body->mbo_t_state & MS_RESTORE)
2759 set_bit(LLIF_FILE_RESTORING, &lli->lli_flags);
2761 clear_bit(LLIF_FILE_RESTORING, &lli->lli_flags);
2767 /* child default LMV is inherited from parent */
2768 static inline bool ll_default_lmv_inherited(struct lmv_stripe_md *pdmv,
2769 struct lmv_stripe_md *cdmv)
2774 if (pdmv->lsm_md_magic != cdmv->lsm_md_magic ||
2775 pdmv->lsm_md_stripe_count != cdmv->lsm_md_stripe_count ||
2776 pdmv->lsm_md_master_mdt_index != cdmv->lsm_md_master_mdt_index ||
2777 pdmv->lsm_md_hash_type != cdmv->lsm_md_hash_type)
2780 if (cdmv->lsm_md_max_inherit !=
2781 lmv_inherit_next(pdmv->lsm_md_max_inherit))
2784 if (cdmv->lsm_md_max_inherit_rr !=
2785 lmv_inherit_rr_next(pdmv->lsm_md_max_inherit_rr))
2791 /* update directory depth to ROOT, called after LOOKUP lock is fetched. */
2792 void ll_update_dir_depth(struct inode *dir, struct inode *inode)
2794 struct ll_inode_info *plli;
2795 struct ll_inode_info *lli;
2797 if (!S_ISDIR(inode->i_mode))
2803 plli = ll_i2info(dir);
2804 lli = ll_i2info(inode);
2805 lli->lli_dir_depth = plli->lli_dir_depth + 1;
2806 if (plli->lli_default_lsm_md && lli->lli_default_lsm_md) {
2807 down_read(&plli->lli_lsm_sem);
2808 down_read(&lli->lli_lsm_sem);
2809 if (ll_default_lmv_inherited(plli->lli_default_lsm_md,
2810 lli->lli_default_lsm_md))
2811 lli->lli_inherit_depth =
2812 plli->lli_inherit_depth + 1;
2814 lli->lli_inherit_depth = 0;
2815 up_read(&lli->lli_lsm_sem);
2816 up_read(&plli->lli_lsm_sem);
2818 lli->lli_inherit_depth = 0;
2821 CDEBUG(D_INODE, DFID" depth %hu default LMV depth %hu\n",
2822 PFID(&lli->lli_fid), lli->lli_dir_depth, lli->lli_inherit_depth);
2825 void ll_truncate_inode_pages_final(struct inode *inode)
2827 struct address_space *mapping = &inode->i_data;
2828 unsigned long nrpages;
2829 unsigned long flags;
2831 truncate_inode_pages_final(mapping);
2833 /* Workaround for LU-118: Note nrpages may not be totally updated when
2834 * truncate_inode_pages() returns, as there can be a page in the process
2835 * of deletion (inside __delete_from_page_cache()) in the specified
2836 * range. Thus mapping->nrpages can be non-zero when this function
2837 * returns even after truncation of the whole mapping. Only do this if
2838 * npages isn't already zero.
2840 nrpages = mapping->nrpages;
2842 ll_xa_lock_irqsave(&mapping->i_pages, flags);
2843 nrpages = mapping->nrpages;
2844 ll_xa_unlock_irqrestore(&mapping->i_pages, flags);
2845 } /* Workaround end */
2847 LASSERTF(nrpages == 0, "%s: inode="DFID"(%p) nrpages=%lu, "
2848 "see https://jira.whamcloud.com/browse/LU-118\n",
2849 ll_i2sbi(inode)->ll_fsname,
2850 PFID(ll_inode2fid(inode)), inode, nrpages);
2853 int ll_read_inode2(struct inode *inode, void *opaque)
2855 struct lustre_md *md = opaque;
2856 struct ll_inode_info *lli = ll_i2info(inode);
2860 CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p)\n",
2861 PFID(&lli->lli_fid), inode);
2863 /* Core attributes from the MDS first. This is a new inode, and
2864 * the VFS doesn't zero times in the core inode so we have to do
2865 * it ourselves. They will be overwritten by either MDS or OST
2866 * attributes - we just need to make sure they aren't newer.
2868 inode->i_mtime.tv_sec = 0;
2869 inode->i_atime.tv_sec = 0;
2870 inode->i_ctime.tv_sec = 0;
2872 rc = ll_update_inode(inode, md);
2876 /* OIDEBUG(inode); */
2878 #ifdef HAVE_BACKING_DEV_INFO
2879 /* initializing backing dev info. */
2880 inode->i_mapping->backing_dev_info = &s2lsi(inode->i_sb)->lsi_bdi;
2882 if (S_ISREG(inode->i_mode)) {
2883 struct ll_sb_info *sbi = ll_i2sbi(inode);
2884 inode->i_op = &ll_file_inode_operations;
2885 inode->i_fop = sbi->ll_fop;
2886 inode->i_mapping->a_ops = (struct address_space_operations *)&ll_aops;
2888 } else if (S_ISDIR(inode->i_mode)) {
2889 inode->i_op = &ll_dir_inode_operations;
2890 inode->i_fop = &ll_dir_operations;
2892 } else if (S_ISLNK(inode->i_mode)) {
2893 inode->i_op = &ll_fast_symlink_inode_operations;
2896 inode->i_op = &ll_special_inode_operations;
2898 init_special_inode(inode, inode->i_mode,
2907 void ll_delete_inode(struct inode *inode)
2909 struct ll_inode_info *lli = ll_i2info(inode);
2912 if (S_ISREG(inode->i_mode) && lli->lli_clob != NULL) {
2913 /* It is last chance to write out dirty pages,
2914 * otherwise we may lose data while umount.
2916 * If i_nlink is 0 then just discard data. This is safe because
2917 * local inode gets i_nlink 0 from server only for the last
2918 * unlink, so that file is not opened somewhere else
2920 cl_sync_file_range(inode, 0, OBD_OBJECT_EOF, inode->i_nlink ?
2921 CL_FSYNC_LOCAL : CL_FSYNC_DISCARD, 1);
2924 ll_truncate_inode_pages_final(inode);
2925 ll_clear_inode(inode);
2931 int ll_iocontrol(struct inode *inode, struct file *file,
2932 unsigned int cmd, unsigned long arg)
2934 struct ll_sb_info *sbi = ll_i2sbi(inode);
2935 struct ptlrpc_request *req = NULL;
2940 case FS_IOC_GETFLAGS: {
2941 struct mdt_body *body;
2942 struct md_op_data *op_data;
2944 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL,
2945 0, 0, LUSTRE_OPC_ANY,
2947 if (IS_ERR(op_data))
2948 RETURN(PTR_ERR(op_data));
2950 op_data->op_valid = OBD_MD_FLFLAGS;
2951 rc = md_getattr(sbi->ll_md_exp, op_data, &req);
2952 ll_finish_md_op_data(op_data);
2954 CERROR("%s: failure inode "DFID": rc = %d\n",
2955 sbi->ll_md_exp->exp_obd->obd_name,
2956 PFID(ll_inode2fid(inode)), rc);
2960 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
2962 flags = body->mbo_flags;
2964 ptlrpc_req_finished(req);
2966 RETURN(put_user(flags, (int __user *)arg));
2968 case FS_IOC_SETFLAGS: {
2970 struct md_op_data *op_data;
2971 struct cl_object *obj;
2972 struct fsxattr fa = { 0 };
2974 if (get_user(flags, (int __user *)arg))
2977 fa.fsx_projid = ll_i2info(inode)->lli_projid;
2978 if (flags & LUSTRE_PROJINHERIT_FL)
2979 fa.fsx_xflags = FS_XFLAG_PROJINHERIT;
2981 rc = ll_ioctl_check_project(inode, fa.fsx_xflags,
2986 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
2987 LUSTRE_OPC_ANY, NULL);
2988 if (IS_ERR(op_data))
2989 RETURN(PTR_ERR(op_data));
2991 op_data->op_attr_flags = flags;
2992 op_data->op_xvalid |= OP_XVALID_FLAGS;
2993 rc = md_setattr(sbi->ll_md_exp, op_data, NULL, 0, &req);
2994 ll_finish_md_op_data(op_data);
2995 ptlrpc_req_finished(req);
2999 ll_update_inode_flags(inode, flags);
3001 obj = ll_i2info(inode)->lli_clob;
3005 OBD_ALLOC_PTR(attr);
3009 rc = cl_setattr_ost(obj, attr, OP_XVALID_FLAGS, flags);
3021 int ll_flush_ctx(struct inode *inode)
3023 struct ll_sb_info *sbi = ll_i2sbi(inode);
3025 CDEBUG(D_SEC, "flush context for user %d\n",
3026 from_kuid(&init_user_ns, current_uid()));
3028 obd_set_info_async(NULL, sbi->ll_md_exp,
3029 sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
3031 obd_set_info_async(NULL, sbi->ll_dt_exp,
3032 sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
3037 /* umount -f client means force down, don't save state */
3038 void ll_umount_begin(struct super_block *sb)
3040 struct ll_sb_info *sbi = ll_s2sbi(sb);
3041 struct obd_device *obd;
3042 struct obd_ioctl_data *ioc_data;
3046 CDEBUG(D_VFSTRACE, "VFS Op: superblock %p count %d active %d\n", sb,
3047 sb->s_count, atomic_read(&sb->s_active));
3049 obd = class_exp2obd(sbi->ll_md_exp);
3051 CERROR("Invalid MDC connection handle %#llx\n",
3052 sbi->ll_md_exp->exp_handle.h_cookie);
3058 obd = class_exp2obd(sbi->ll_dt_exp);
3060 CERROR("Invalid LOV connection handle %#llx\n",
3061 sbi->ll_dt_exp->exp_handle.h_cookie);
3067 OBD_ALLOC_PTR(ioc_data);
3069 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_md_exp,
3070 sizeof *ioc_data, ioc_data, NULL);
3072 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_dt_exp,
3073 sizeof *ioc_data, ioc_data, NULL);
3075 OBD_FREE_PTR(ioc_data);
3078 /* Really, we'd like to wait until there are no requests outstanding,
3079 * and then continue. For now, we just periodically checking for vfs
3080 * to decrement mnt_cnt and hope to finish it within 10sec.
3084 !may_umount(sbi->ll_mnt.mnt)) {
3092 int ll_remount_fs(struct super_block *sb, int *flags, char *data)
3094 struct ll_sb_info *sbi = ll_s2sbi(sb);
3095 char *profilenm = get_profile_name(sb);
3099 if ((*flags & MS_RDONLY) != (sb->s_flags & SB_RDONLY)) {
3100 read_only = *flags & MS_RDONLY;
3101 err = obd_set_info_async(NULL, sbi->ll_md_exp,
3102 sizeof(KEY_READ_ONLY),
3103 KEY_READ_ONLY, sizeof(read_only),
3106 LCONSOLE_WARN("Failed to remount %s %s (%d)\n",
3107 profilenm, read_only ?
3108 "read-only" : "read-write", err);
3113 sb->s_flags |= SB_RDONLY;
3115 sb->s_flags &= ~SB_RDONLY;
3117 if (test_bit(LL_SBI_VERBOSE, sbi->ll_flags))
3118 LCONSOLE_WARN("Remounted %s %s\n", profilenm,
3119 read_only ? "read-only" : "read-write");
3125 * Cleanup the open handle that is cached on MDT-side.
3127 * For open case, the client side open handling thread may hit error
3128 * after the MDT grant the open. Under such case, the client should
3129 * send close RPC to the MDT as cleanup; otherwise, the open handle
3130 * on the MDT will be leaked there until the client umount or evicted.
3132 * In further, if someone unlinked the file, because the open handle
3133 * holds the reference on such file/object, then it will block the
3134 * subsequent threads that want to locate such object via FID.
3136 * \param[in] sb super block for this file-system
3137 * \param[in] open_req pointer to the original open request
3139 void ll_open_cleanup(struct super_block *sb, struct req_capsule *pill)
3141 struct mdt_body *body;
3142 struct md_op_data *op_data;
3143 struct ptlrpc_request *close_req = NULL;
3144 struct obd_export *exp = ll_s2sbi(sb)->ll_md_exp;
3147 body = req_capsule_server_get(pill, &RMF_MDT_BODY);
3148 OBD_ALLOC_PTR(op_data);
3149 if (op_data == NULL) {
3150 CWARN("%s: cannot allocate op_data to release open handle for "
3151 DFID"\n", ll_s2sbi(sb)->ll_fsname, PFID(&body->mbo_fid1));
3156 op_data->op_fid1 = body->mbo_fid1;
3157 op_data->op_open_handle = body->mbo_open_handle;
3158 op_data->op_mod_time = ktime_get_real_seconds();
3159 md_close(exp, op_data, NULL, &close_req);
3160 ptlrpc_req_finished(close_req);
3161 ll_finish_md_op_data(op_data);
3166 /* set filesystem-wide default LMV for subdir mount if it's enabled on ROOT. */
3167 static int ll_fileset_default_lmv_fixup(struct inode *inode,
3168 struct lustre_md *md)
3170 struct ll_sb_info *sbi = ll_i2sbi(inode);
3171 struct ptlrpc_request *req = NULL;
3172 union lmv_mds_md *lmm = NULL;
3176 LASSERT(is_root_inode(inode));
3177 LASSERT(!fid_is_root(&sbi->ll_root_fid));
3178 LASSERT(!md->default_lmv);
3180 rc = ll_dir_get_default_layout(inode, (void **)&lmm, &size, &req,
3182 GET_DEFAULT_LAYOUT_ROOT);
3183 if (rc && rc != -ENODATA)
3188 rc = md_unpackmd(sbi->ll_md_exp, &md->default_lmv, lmm, size);
3197 ptlrpc_req_finished(req);
3201 int ll_prep_inode(struct inode **inode, struct req_capsule *pill,
3202 struct super_block *sb, struct lookup_intent *it)
3204 struct ll_sb_info *sbi = NULL;
3205 struct lustre_md md = { NULL };
3206 bool default_lmv_deleted = false;
3211 LASSERT(*inode || sb);
3212 sbi = sb ? ll_s2sbi(sb) : ll_i2sbi(*inode);
3213 rc = md_get_lustre_md(sbi->ll_md_exp, pill, sbi->ll_dt_exp,
3214 sbi->ll_md_exp, &md);
3218 if (S_ISDIR(md.body->mbo_mode) && md.lmv && lmv_dir_striped(md.lmv) &&
3219 it && it->it_extra_rpc_check) {
3220 /* TODO: Check @lsm unchanged via @lsm_md_eq. */
3221 it->it_extra_rpc_need = 1;
3222 GOTO(out, rc = -EAGAIN);
3226 * clear default_lmv only if intent_getattr reply doesn't contain it.
3227 * but it needs to be done after iget, check this early because
3228 * ll_update_lsm_md() may change md.
3230 if (it && (it->it_op & (IT_LOOKUP | IT_GETATTR)) &&
3231 S_ISDIR(md.body->mbo_mode) && !md.default_lmv) {
3232 if (unlikely(*inode && is_root_inode(*inode) &&
3233 !fid_is_root(&sbi->ll_root_fid))) {
3234 rc = ll_fileset_default_lmv_fixup(*inode, &md);
3239 if (!md.default_lmv)
3240 default_lmv_deleted = true;
3244 rc = ll_update_inode(*inode, &md);
3248 bool api32 = test_bit(LL_SBI_32BIT_API, sbi->ll_flags);
3249 struct lu_fid *fid1 = &md.body->mbo_fid1;
3251 LASSERT(sb != NULL);
3254 * At this point server returns to client's same fid as client
3255 * generated for creating. So using ->fid1 is okay here.
3257 if (!fid_is_sane(fid1)) {
3258 CERROR("%s: Fid is insane "DFID"\n",
3259 sbi->ll_fsname, PFID(fid1));
3260 GOTO(out, rc = -EINVAL);
3263 *inode = ll_iget(sb, cl_fid_build_ino(fid1, api32), &md);
3264 if (IS_ERR(*inode)) {
3266 rc = IS_ERR(*inode) ? PTR_ERR(*inode) : -ENOMEM;
3268 CERROR("new_inode -fatal: rc %d\n", rc);
3273 /* Handling piggyback layout lock.
3274 * Layout lock can be piggybacked by getattr and open request.
3275 * The lsm can be applied to inode only if it comes with a layout lock
3276 * otherwise correct layout may be overwritten, for example:
3277 * 1. proc1: mdt returns a lsm but not granting layout
3278 * 2. layout was changed by another client
3279 * 3. proc2: refresh layout and layout lock granted
3280 * 4. proc1: to apply a stale layout */
3281 if (it != NULL && it->it_lock_mode != 0) {
3282 struct lustre_handle lockh;
3283 struct ldlm_lock *lock;
3285 lockh.cookie = it->it_lock_handle;
3286 lock = ldlm_handle2lock(&lockh);
3287 LASSERT(lock != NULL);
3288 if (ldlm_has_layout(lock)) {
3289 struct cl_object_conf conf;
3291 memset(&conf, 0, sizeof(conf));
3292 conf.coc_opc = OBJECT_CONF_SET;
3293 conf.coc_inode = *inode;
3294 conf.coc_lock = lock;
3295 conf.u.coc_layout = md.layout;
3296 (void)ll_layout_conf(*inode, &conf);
3298 LDLM_LOCK_PUT(lock);
3301 if (default_lmv_deleted)
3302 ll_update_default_lsm_md(*inode, &md);
3304 /* we may want to apply some policy for foreign file/dir */
3305 if (ll_sbi_has_foreign_symlink(sbi)) {
3306 rc = ll_manage_foreign(*inode, &md);
3314 /* cleanup will be done if necessary */
3315 md_free_lustre_md(sbi->ll_md_exp, &md);
3317 if (rc != 0 && it != NULL && it->it_op & IT_OPEN) {
3318 ll_intent_drop_lock(it);
3319 ll_open_cleanup(sb != NULL ? sb : (*inode)->i_sb, pill);
3325 int ll_obd_statfs(struct inode *inode, void __user *arg)
3327 struct ll_sb_info *sbi = NULL;
3328 struct obd_export *exp;
3329 struct obd_ioctl_data *data = NULL;
3334 sbi = ll_i2sbi(inode);
3336 GOTO(out_statfs, rc = -EINVAL);
3338 rc = obd_ioctl_getdata(&data, &len, arg);
3340 GOTO(out_statfs, rc);
3342 if (!data->ioc_inlbuf1 || !data->ioc_inlbuf2 ||
3343 !data->ioc_pbuf1 || !data->ioc_pbuf2)
3344 GOTO(out_statfs, rc = -EINVAL);
3346 if (data->ioc_inllen1 != sizeof(__u32) ||
3347 data->ioc_inllen2 != sizeof(__u32) ||
3348 data->ioc_plen1 != sizeof(struct obd_statfs) ||
3349 data->ioc_plen2 != sizeof(struct obd_uuid))
3350 GOTO(out_statfs, rc = -EINVAL);
3352 memcpy(&type, data->ioc_inlbuf1, sizeof(__u32));
3353 if (type & LL_STATFS_LMV)
3354 exp = sbi->ll_md_exp;
3355 else if (type & LL_STATFS_LOV)
3356 exp = sbi->ll_dt_exp;
3358 GOTO(out_statfs, rc = -ENODEV);
3360 rc = obd_iocontrol(IOC_OBD_STATFS, exp, len, data, NULL);
3362 GOTO(out_statfs, rc);
3364 OBD_FREE_LARGE(data, len);
3369 * this is normally called in ll_fini_md_op_data(), but sometimes it needs to
3370 * be called early to avoid deadlock.
3372 void ll_unlock_md_op_lsm(struct md_op_data *op_data)
3374 if (op_data->op_mea2_sem) {
3375 up_read_non_owner(op_data->op_mea2_sem);
3376 op_data->op_mea2_sem = NULL;
3379 if (op_data->op_mea1_sem) {
3380 up_read_non_owner(op_data->op_mea1_sem);
3381 op_data->op_mea1_sem = NULL;
3385 /* this function prepares md_op_data hint for passing it down to MD stack. */
3386 struct md_op_data *ll_prep_md_op_data(struct md_op_data *op_data,
3387 struct inode *i1, struct inode *i2,
3388 const char *name, size_t namelen,
3389 __u32 mode, enum md_op_code opc,
3392 struct llcrypt_name fname = { 0 };
3395 LASSERT(i1 != NULL);
3398 /* Do not reuse namelen for something else. */
3400 return ERR_PTR(-EINVAL);
3402 if ((!IS_ENCRYPTED(i1) ||
3403 (opc != LUSTRE_OPC_LOOKUP && opc != LUSTRE_OPC_CREATE)) &&
3404 namelen > ll_i2sbi(i1)->ll_namelen)
3405 return ERR_PTR(-ENAMETOOLONG);
3407 /* "/" is not valid name, but it's allowed */
3408 if (!lu_name_is_valid_2(name, namelen) &&
3409 strncmp("/", name, namelen) != 0)
3410 return ERR_PTR(-EINVAL);
3413 if (op_data == NULL)
3414 OBD_ALLOC_PTR(op_data);
3416 if (op_data == NULL)
3417 return ERR_PTR(-ENOMEM);
3419 ll_i2gids(op_data->op_suppgids, i1, i2);
3420 /* If the client is using a subdir mount and looks at what it sees as
3421 * /.fscrypt, interpret it as the .fscrypt dir at the root of the fs.
3423 if (unlikely(i1->i_sb && i1->i_sb->s_root && is_root_inode(i1) &&
3424 !fid_is_root(ll_inode2fid(i1)) &&
3425 name && namelen == strlen(dot_fscrypt_name) &&
3426 strncmp(name, dot_fscrypt_name, namelen) == 0))
3427 lu_root_fid(&op_data->op_fid1);
3429 op_data->op_fid1 = *ll_inode2fid(i1);
3431 if (S_ISDIR(i1->i_mode)) {
3432 down_read_non_owner(&ll_i2info(i1)->lli_lsm_sem);
3433 op_data->op_mea1_sem = &ll_i2info(i1)->lli_lsm_sem;
3434 op_data->op_mea1 = ll_i2info(i1)->lli_lsm_md;
3435 op_data->op_default_mea1 = ll_i2info(i1)->lli_default_lsm_md;
3439 op_data->op_fid2 = *ll_inode2fid(i2);
3440 if (S_ISDIR(i2->i_mode)) {
3442 /* i2 is typically a child of i1, and MUST be
3443 * further from the root to avoid deadlocks.
3445 down_read_non_owner(&ll_i2info(i2)->lli_lsm_sem);
3446 op_data->op_mea2_sem =
3447 &ll_i2info(i2)->lli_lsm_sem;
3449 op_data->op_mea2 = ll_i2info(i2)->lli_lsm_md;
3452 fid_zero(&op_data->op_fid2);
3455 if (test_bit(LL_SBI_64BIT_HASH, ll_i2sbi(i1)->ll_flags))
3456 op_data->op_cli_flags |= CLI_HASH64;
3458 if (ll_need_32bit_api(ll_i2sbi(i1)))
3459 op_data->op_cli_flags |= CLI_API32;
3461 if ((i2 && is_root_inode(i2)) ||
3462 opc == LUSTRE_OPC_LOOKUP || opc == LUSTRE_OPC_CREATE) {
3463 /* In case of lookup, ll_setup_filename() has already been
3464 * called in ll_lookup_it(), so just take provided name.
3465 * Also take provided name if we are dealing with root inode.
3467 fname.disk_name.name = (unsigned char *)name;
3468 fname.disk_name.len = namelen;
3469 } else if (name && namelen) {
3470 struct qstr dname = QSTR_INIT(name, namelen);
3472 struct lu_fid *pfid = NULL;
3476 if (!S_ISDIR(i1->i_mode) && i2 && S_ISDIR(i2->i_mode)) {
3477 /* special case when called from ll_link() */
3482 lookup = (int)(opc == LUSTRE_OPC_ANY);
3484 if (opc == LUSTRE_OPC_ANY && lookup)
3486 rc = ll_setup_filename(dir, &dname, lookup, &fname, pfid);
3488 ll_finish_md_op_data(op_data);
3491 if (pfid && !fid_is_zero(pfid)) {
3493 op_data->op_fid2 = fid;
3494 op_data->op_bias = MDS_FID_OP;
3496 if (fname.disk_name.name &&
3497 fname.disk_name.name != (unsigned char *)name) {
3498 /* op_data->op_name must be freed after use */
3499 op_data->op_flags |= MF_OPNAME_KMALLOCED;
3503 /* In fact LUSTRE_OPC_LOOKUP, LUSTRE_OPC_OPEN
3504 * are LUSTRE_OPC_ANY
3506 if (opc == LUSTRE_OPC_LOOKUP || opc == LUSTRE_OPC_OPEN)
3507 op_data->op_code = LUSTRE_OPC_ANY;
3509 op_data->op_code = opc;
3510 op_data->op_name = fname.disk_name.name;
3511 op_data->op_namelen = fname.disk_name.len;
3512 op_data->op_mode = mode;
3513 op_data->op_mod_time = ktime_get_real_seconds();
3514 op_data->op_fsuid = from_kuid(&init_user_ns, current_fsuid());
3515 op_data->op_fsgid = from_kgid(&init_user_ns, current_fsgid());
3516 op_data->op_cap = current_cap();
3517 op_data->op_mds = 0;
3518 if ((opc == LUSTRE_OPC_CREATE) && (name != NULL) &&
3519 filename_is_volatile(name, namelen, &op_data->op_mds)) {
3520 op_data->op_bias |= MDS_CREATE_VOLATILE;
3522 op_data->op_data = data;
3527 void ll_finish_md_op_data(struct md_op_data *op_data)
3529 ll_unlock_md_op_lsm(op_data);
3530 ll_security_release_secctx(op_data->op_file_secctx,
3531 op_data->op_file_secctx_size);
3532 if (op_data->op_flags & MF_OPNAME_KMALLOCED)
3533 /* allocated via ll_setup_filename called
3534 * from ll_prep_md_op_data
3536 kfree(op_data->op_name);
3537 llcrypt_free_ctx(op_data->op_file_encctx, op_data->op_file_encctx_size);
3538 OBD_FREE_PTR(op_data);
3541 int ll_show_options(struct seq_file *seq, struct dentry *dentry)
3543 struct ll_sb_info *sbi;
3546 LASSERT(seq && dentry);
3547 sbi = ll_s2sbi(dentry->d_sb);
3549 if (test_bit(LL_SBI_NOLCK, sbi->ll_flags))
3550 seq_puts(seq, "nolock");
3552 for (i = 1; ll_sbi_flags_name[i].token != LL_SBI_NUM_MOUNT_OPT; i++) {
3553 /* match_table in some cases has patterns for both enabled and
3554 * disabled cases. Ignore 'no'xxx versions if bit is set.
3556 if (test_bit(ll_sbi_flags_name[i].token, sbi->ll_flags) &&
3557 strncmp(ll_sbi_flags_name[i].pattern, "no", 2)) {
3558 if (ll_sbi_flags_name[i].token ==
3559 LL_SBI_FOREIGN_SYMLINK) {
3560 seq_show_option(seq, "foreign_symlink",
3561 sbi->ll_foreign_symlink_prefix);
3563 seq_printf(seq, ",%s",
3564 ll_sbi_flags_name[i].pattern);
3567 /* You can have either localflock or flock but not
3568 * both. If localflock is set don't print flock or
3571 if (ll_sbi_flags_name[i].token == LL_SBI_LOCALFLOCK)
3573 } else if (!test_bit(ll_sbi_flags_name[i].token, sbi->ll_flags) &&
3574 !strncmp(ll_sbi_flags_name[i].pattern, "no", 2)) {
3575 seq_printf(seq, ",%s",
3576 ll_sbi_flags_name[i].pattern);
3580 llcrypt_show_test_dummy_encryption(seq, ',', dentry->d_sb);
3586 * Get obd name by cmd, and copy out to user space
3588 int ll_get_obd_name(struct inode *inode, unsigned int cmd, unsigned long arg)
3590 struct ll_sb_info *sbi = ll_i2sbi(inode);
3591 struct obd_device *obd;
3594 if (cmd == OBD_IOC_GETNAME_OLD || cmd == OBD_IOC_GETDTNAME)
3595 obd = class_exp2obd(sbi->ll_dt_exp);
3596 else if (cmd == OBD_IOC_GETMDNAME)
3597 obd = class_exp2obd(sbi->ll_md_exp);
3604 if (copy_to_user((void __user *)arg, obd->obd_name,
3605 strlen(obd->obd_name) + 1))
3612 struct work_struct db_work;
3613 struct dentry *db_dentry;
3614 /* Let's hope the path is not too long, 32 bytes for the work struct
3617 char buf[PAGE_SIZE - sizeof(struct work_struct) - sizeof(void *)];
3620 static void ll_dput_later(struct work_struct *work)
3622 struct dname_buf *db = container_of(work, struct dname_buf, db_work);
3624 dput(db->db_dentry);
3625 free_page((unsigned long)db);
3628 static char* ll_d_path(struct dentry *dentry, char *buf, int bufsize)
3635 p.mnt = current->fs->root.mnt;
3637 path = d_path(&p, buf, bufsize);
3642 void ll_dirty_page_discard_warn(struct inode *inode, int ioret)
3644 struct dname_buf *db;
3646 struct dentry *dentry = NULL;
3648 /* this can be called inside spin lock so use GFP_ATOMIC. */
3649 db = (struct dname_buf *)__get_free_page(GFP_ATOMIC);
3652 dentry = d_find_alias(inode);
3654 path = ll_d_path(dentry, db->buf, sizeof(db->buf));
3657 /* The below message is checked in recovery-small.sh test_24b */
3659 "%s: dirty page discard: %s/fid: "DFID"/%s may get corrupted "
3660 "(rc %d)\n", ll_i2sbi(inode)->ll_fsname,
3661 s2lsi(inode->i_sb)->lsi_lmd->lmd_dev,
3662 PFID(ll_inode2fid(inode)),
3663 (path && !IS_ERR(path)) ? path : "", ioret);
3665 if (dentry != NULL) {
3666 /* We cannot dput here since if we happen to be the last holder
3667 * then we can end up waiting for page evictions that
3668 * in turn wait for RPCs that need this instance of ptlrpcd
3669 * (callng brw_interpret->*page_completion*->vmpage_error->here)
3672 INIT_WORK(&db->db_work, ll_dput_later);
3673 db->db_dentry = dentry;
3674 schedule_work(&db->db_work);
3677 free_page((unsigned long)db);
3681 ssize_t ll_copy_user_md(const struct lov_user_md __user *md,
3682 struct lov_user_md **kbuf)
3684 struct lov_user_md lum;
3688 if (copy_from_user(&lum, md, sizeof(lum)))
3691 lum_size = ll_lov_user_md_size(&lum);
3695 OBD_ALLOC_LARGE(*kbuf, lum_size);
3699 if (copy_from_user(*kbuf, md, lum_size) != 0) {
3700 OBD_FREE_LARGE(*kbuf, lum_size);
3708 * Compute llite root squash state after a change of root squash
3709 * configuration setting or add/remove of a lnet nid
3711 void ll_compute_rootsquash_state(struct ll_sb_info *sbi)
3713 struct root_squash_info *squash = &sbi->ll_squash;
3716 struct lnet_processid id;
3718 /* Update norootsquash flag */
3719 spin_lock(&squash->rsi_lock);
3720 if (list_empty(&squash->rsi_nosquash_nids))
3721 clear_bit(LL_SBI_NOROOTSQUASH, sbi->ll_flags);
3723 /* Do not apply root squash as soon as one of our NIDs is
3724 * in the nosquash_nids list */
3727 while (LNetGetId(i++, &id) != -ENOENT) {
3728 if (nid_is_lo0(&id.nid))
3730 if (cfs_match_nid(lnet_nid_to_nid4(&id.nid),
3731 &squash->rsi_nosquash_nids)) {
3737 set_bit(LL_SBI_NOROOTSQUASH, sbi->ll_flags);
3739 clear_bit(LL_SBI_NOROOTSQUASH, sbi->ll_flags);
3741 spin_unlock(&squash->rsi_lock);
3745 * Parse linkea content to extract information about a given hardlink
3747 * \param[in] ldata - Initialized linkea data
3748 * \param[in] linkno - Link identifier
3749 * \param[out] parent_fid - The entry's parent FID
3750 * \param[out] ln - Entry name destination buffer
3752 * \retval 0 on success
3753 * \retval Appropriate negative error code on failure
3755 static int ll_linkea_decode(struct linkea_data *ldata, unsigned int linkno,
3756 struct lu_fid *parent_fid, struct lu_name *ln)
3762 rc = linkea_init_with_rec(ldata);
3766 if (linkno >= ldata->ld_leh->leh_reccount)
3767 /* beyond last link */
3770 linkea_first_entry(ldata);
3771 for (idx = 0; ldata->ld_lee != NULL; idx++) {
3772 linkea_entry_unpack(ldata->ld_lee, &ldata->ld_reclen, ln,
3777 linkea_next_entry(ldata);
3787 * Get parent FID and name of an identified link. Operation is performed for
3788 * a given link number, letting the caller iterate over linkno to list one or
3789 * all links of an entry.
3791 * \param[in] file - File descriptor against which to perform the operation
3792 * \param[in,out] arg - User-filled structure containing the linkno to operate
3793 * on and the available size. It is eventually filled with
3794 * the requested information or left untouched on error
3796 * \retval - 0 on success
3797 * \retval - Appropriate negative error code on failure
3799 int ll_getparent(struct file *file, struct getparent __user *arg)
3801 struct inode *inode = file_inode(file);
3802 struct linkea_data *ldata;
3803 struct lu_buf buf = LU_BUF_NULL;
3805 struct lu_fid parent_fid;
3812 if (!capable(CAP_DAC_READ_SEARCH) &&
3813 !test_bit(LL_SBI_USER_FID2PATH, ll_i2sbi(inode)->ll_flags))
3816 if (get_user(name_size, &arg->gp_name_size))
3819 if (get_user(linkno, &arg->gp_linkno))
3822 if (name_size > PATH_MAX)
3825 OBD_ALLOC(ldata, sizeof(*ldata));
3829 rc = linkea_data_new(ldata, &buf);
3831 GOTO(ldata_free, rc);
3833 rc = ll_xattr_list(inode, XATTR_NAME_LINK, XATTR_TRUSTED_T, buf.lb_buf,
3834 buf.lb_len, OBD_MD_FLXATTR);
3838 rc = ll_linkea_decode(ldata, linkno, &parent_fid, &ln);
3842 if (ln.ln_namelen >= name_size)
3843 GOTO(lb_free, rc = -EOVERFLOW);
3845 if (copy_to_user(&arg->gp_fid, &parent_fid, sizeof(arg->gp_fid)))
3846 GOTO(lb_free, rc = -EFAULT);
3848 if (copy_to_user(&arg->gp_name, ln.ln_name, ln.ln_namelen))
3849 GOTO(lb_free, rc = -EFAULT);
3851 if (put_user('\0', arg->gp_name + ln.ln_namelen))
3852 GOTO(lb_free, rc = -EFAULT);
3857 OBD_FREE(ldata, sizeof(*ldata));