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 set_bit(LL_SBI_AGL_ENABLED, sbi->ll_flags);
182 set_bit(LL_SBI_FAST_READ, sbi->ll_flags);
183 set_bit(LL_SBI_TINY_WRITE, sbi->ll_flags);
184 set_bit(LL_SBI_PARALLEL_DIO, sbi->ll_flags);
185 ll_sbi_set_encrypt(sbi, true);
186 ll_sbi_set_name_encrypt(sbi, true);
189 sbi->ll_squash.rsi_uid = 0;
190 sbi->ll_squash.rsi_gid = 0;
191 INIT_LIST_HEAD(&sbi->ll_squash.rsi_nosquash_nids);
192 spin_lock_init(&sbi->ll_squash.rsi_lock);
194 /* Per-filesystem file heat */
195 sbi->ll_heat_decay_weight = SBI_DEFAULT_HEAT_DECAY_WEIGHT;
196 sbi->ll_heat_period_second = SBI_DEFAULT_HEAT_PERIOD_SECOND;
198 /* Per-fs open heat level before requesting open lock */
199 sbi->ll_oc_thrsh_count = SBI_DEFAULT_OPENCACHE_THRESHOLD_COUNT;
200 sbi->ll_oc_max_ms = SBI_DEFAULT_OPENCACHE_THRESHOLD_MAX_MS;
201 sbi->ll_oc_thrsh_ms = SBI_DEFAULT_OPENCACHE_THRESHOLD_MS;
204 if (sbi->ll_foreign_symlink_prefix)
205 OBD_FREE(sbi->ll_foreign_symlink_prefix, sizeof("/mnt/"));
207 cl_cache_decref(sbi->ll_cache);
208 sbi->ll_cache = NULL;
210 destroy_workqueue(sbi->ll_ra_info.ll_readahead_wq);
212 pcc_super_fini(&sbi->ll_pcc_super);
218 static void ll_free_sbi(struct super_block *sb)
220 struct ll_sb_info *sbi = ll_s2sbi(sb);
224 if (!list_empty(&sbi->ll_squash.rsi_nosquash_nids))
225 cfs_free_nidlist(&sbi->ll_squash.rsi_nosquash_nids);
226 if (sbi->ll_ra_info.ll_readahead_wq)
227 destroy_workqueue(sbi->ll_ra_info.ll_readahead_wq);
228 if (sbi->ll_cache != NULL) {
229 cl_cache_decref(sbi->ll_cache);
230 sbi->ll_cache = NULL;
232 if (sbi->ll_foreign_symlink_prefix) {
233 OBD_FREE(sbi->ll_foreign_symlink_prefix,
234 sbi->ll_foreign_symlink_prefix_size);
235 sbi->ll_foreign_symlink_prefix = NULL;
237 if (sbi->ll_foreign_symlink_upcall) {
238 OBD_FREE(sbi->ll_foreign_symlink_upcall,
239 strlen(sbi->ll_foreign_symlink_upcall) +
241 sbi->ll_foreign_symlink_upcall = NULL;
243 if (sbi->ll_foreign_symlink_upcall_items) {
245 int nb_items = sbi->ll_foreign_symlink_upcall_nb_items;
246 struct ll_foreign_symlink_upcall_item *items =
247 sbi->ll_foreign_symlink_upcall_items;
249 for (i = 0 ; i < nb_items; i++)
250 if (items[i].type == STRING_TYPE)
251 OBD_FREE(items[i].string,
254 OBD_FREE_LARGE(items, nb_items *
255 sizeof(struct ll_foreign_symlink_upcall_item));
256 sbi->ll_foreign_symlink_upcall_items = NULL;
258 ll_free_rw_stats_info(sbi);
259 pcc_super_fini(&sbi->ll_pcc_super);
260 OBD_FREE(sbi, sizeof(*sbi));
265 static int client_common_fill_super(struct super_block *sb, char *md, char *dt)
267 struct inode *root = NULL;
268 struct ll_sb_info *sbi = ll_s2sbi(sb);
269 struct obd_statfs *osfs = NULL;
270 struct ptlrpc_request *request = NULL;
271 struct obd_connect_data *data = NULL;
272 struct obd_uuid *uuid;
273 struct md_op_data *op_data;
274 struct lustre_md lmd;
276 int size, err, checksum;
282 sbi->ll_md_obd = class_name2obd(md);
283 if (!sbi->ll_md_obd) {
284 CERROR("MD %s: not setup or attached\n", md);
298 /* pass client page size via ocd_grant_blkbits, the server should report
299 * back its backend blocksize for grant calculation purpose */
300 data->ocd_grant_blkbits = PAGE_SHIFT;
302 /* indicate MDT features supported by this client */
303 data->ocd_connect_flags = OBD_CONNECT_IBITS | OBD_CONNECT_NODEVOH |
304 OBD_CONNECT_ATTRFID | OBD_CONNECT_GRANT |
305 OBD_CONNECT_VERSION | OBD_CONNECT_BRW_SIZE |
306 OBD_CONNECT_SRVLOCK |
307 OBD_CONNECT_MDS_CAPA | OBD_CONNECT_OSS_CAPA |
308 OBD_CONNECT_CANCELSET | OBD_CONNECT_FID |
309 OBD_CONNECT_AT | OBD_CONNECT_LOV_V3 |
310 OBD_CONNECT_VBR | OBD_CONNECT_FULL20 |
311 OBD_CONNECT_64BITHASH |
312 OBD_CONNECT_EINPROGRESS |
313 OBD_CONNECT_JOBSTATS | OBD_CONNECT_LVB_TYPE |
314 OBD_CONNECT_LAYOUTLOCK | OBD_CONNECT_PINGLESS|
315 OBD_CONNECT_MAX_EASIZE |
316 OBD_CONNECT_FLOCK_DEAD |
317 OBD_CONNECT_DISP_STRIPE | OBD_CONNECT_LFSCK |
318 OBD_CONNECT_OPEN_BY_FID |
319 OBD_CONNECT_DIR_STRIPE |
320 OBD_CONNECT_BULK_MBITS | OBD_CONNECT_CKSUM |
321 OBD_CONNECT_SUBTREE |
322 OBD_CONNECT_MULTIMODRPCS |
323 OBD_CONNECT_GRANT_PARAM |
324 OBD_CONNECT_GRANT_SHRINK |
325 OBD_CONNECT_SHORTIO | OBD_CONNECT_FLAGS2;
327 data->ocd_connect_flags2 = OBD_CONNECT2_DIR_MIGRATE |
328 OBD_CONNECT2_SUM_STATFS |
329 OBD_CONNECT2_OVERSTRIPING |
331 OBD_CONNECT2_LOCK_CONVERT |
332 OBD_CONNECT2_ARCHIVE_ID_ARRAY |
333 OBD_CONNECT2_INC_XID |
335 OBD_CONNECT2_ASYNC_DISCARD |
337 OBD_CONNECT2_CRUSH | OBD_CONNECT2_LSEEK |
338 OBD_CONNECT2_GETATTR_PFID |
339 OBD_CONNECT2_DOM_LVB |
340 OBD_CONNECT2_REP_MBITS |
341 OBD_CONNECT2_ATOMIC_OPEN_LOCK;
343 #ifdef HAVE_LRU_RESIZE_SUPPORT
344 if (test_bit(LL_SBI_LRU_RESIZE, sbi->ll_flags))
345 data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
347 data->ocd_connect_flags |= OBD_CONNECT_ACL_FLAGS;
349 data->ocd_cksum_types = obd_cksum_types_supported_client();
351 if (OBD_FAIL_CHECK(OBD_FAIL_MDC_LIGHTWEIGHT))
352 /* flag mdc connection as lightweight, only used for test
353 * purpose, use with care */
354 data->ocd_connect_flags |= OBD_CONNECT_LIGHTWEIGHT;
356 data->ocd_ibits_known = MDS_INODELOCK_FULL;
357 data->ocd_version = LUSTRE_VERSION_CODE;
359 if (sb->s_flags & SB_RDONLY)
360 data->ocd_connect_flags |= OBD_CONNECT_RDONLY;
361 if (test_bit(LL_SBI_USER_XATTR, sbi->ll_flags))
362 data->ocd_connect_flags |= OBD_CONNECT_XATTR;
365 /* Setting this indicates we correctly support S_NOSEC (See kernel
366 * commit 9e1f1de02c2275d7172e18dc4e7c2065777611bf)
368 sb->s_flags |= SB_NOSEC;
370 sbi->ll_fop = ll_select_file_operations(sbi);
372 /* always ping even if server suppress_pings */
373 if (test_bit(LL_SBI_ALWAYS_PING, sbi->ll_flags))
374 data->ocd_connect_flags &= ~OBD_CONNECT_PINGLESS;
376 obd_connect_set_secctx(data);
377 if (ll_sbi_has_encrypt(sbi)) {
378 obd_connect_set_name_enc(data);
379 obd_connect_set_enc(data);
382 #if defined(CONFIG_SECURITY)
383 data->ocd_connect_flags2 |= OBD_CONNECT2_SELINUX_POLICY;
386 data->ocd_brw_size = MD_MAX_BRW_SIZE;
388 err = obd_connect(NULL, &sbi->ll_md_exp, sbi->ll_md_obd,
389 &sbi->ll_sb_uuid, data, sbi->ll_cache);
391 LCONSOLE_ERROR_MSG(0x14f, "An MDT (md %s) is performing "
392 "recovery, of which this client is not a "
393 "part. Please wait for recovery to complete,"
394 " abort, or time out.\n", md);
397 CERROR("cannot connect to %s: rc = %d\n", md, err);
401 sbi->ll_md_exp->exp_connect_data = *data;
403 err = obd_fid_init(sbi->ll_md_exp->exp_obd, sbi->ll_md_exp,
404 LUSTRE_SEQ_METADATA);
406 CERROR("%s: Can't init metadata layer FID infrastructure, "
407 "rc = %d\n", sbi->ll_md_exp->exp_obd->obd_name, err);
411 /* For mount, we only need fs info from MDT0, and also in DNE, it
412 * can make sure the client can be mounted as long as MDT0 is
414 err = obd_statfs(NULL, sbi->ll_md_exp, osfs,
415 ktime_get_seconds() - sbi->ll_statfs_max_age,
416 OBD_STATFS_FOR_MDT0);
418 GOTO(out_md_fid, err);
420 /* This needs to be after statfs to ensure connect has finished.
421 * Note that "data" does NOT contain the valid connect reply.
422 * If connecting to a 1.8 server there will be no LMV device, so
423 * we can access the MDC export directly and exp_connect_flags will
424 * be non-zero, but if accessing an upgraded 2.1 server it will
425 * have the correct flags filled in.
426 * XXX: fill in the LMV exp_connect_flags from MDC(s). */
427 valid = exp_connect_flags(sbi->ll_md_exp) & CLIENT_CONNECT_MDT_REQD;
428 if (exp_connect_flags(sbi->ll_md_exp) != 0 &&
429 valid != CLIENT_CONNECT_MDT_REQD) {
432 OBD_ALLOC_WAIT(buf, PAGE_SIZE);
433 obd_connect_flags2str(buf, PAGE_SIZE,
434 valid ^ CLIENT_CONNECT_MDT_REQD, 0, ",");
435 LCONSOLE_ERROR_MSG(0x170, "Server %s does not support "
436 "feature(s) needed for correct operation "
437 "of this client (%s). Please upgrade "
438 "server or downgrade client.\n",
439 sbi->ll_md_exp->exp_obd->obd_name, buf);
440 OBD_FREE(buf, PAGE_SIZE);
441 GOTO(out_md_fid, err = -EPROTO);
444 size = sizeof(*data);
445 err = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_CONN_DATA),
446 KEY_CONN_DATA, &size, data);
448 CERROR("%s: Get connect data failed: rc = %d\n",
449 sbi->ll_md_exp->exp_obd->obd_name, err);
450 GOTO(out_md_fid, err);
453 LASSERT(osfs->os_bsize);
454 sb->s_blocksize = osfs->os_bsize;
455 sb->s_blocksize_bits = log2(osfs->os_bsize);
456 sb->s_magic = LL_SUPER_MAGIC;
457 sb->s_maxbytes = MAX_LFS_FILESIZE;
458 sbi->ll_namelen = osfs->os_namelen;
459 sbi->ll_mnt.mnt = current->fs->root.mnt;
460 sbi->ll_mnt_ns = current->nsproxy->mnt_ns;
462 if (test_bit(LL_SBI_USER_XATTR, sbi->ll_flags) &&
463 !(data->ocd_connect_flags & OBD_CONNECT_XATTR)) {
464 LCONSOLE_INFO("Disabling user_xattr feature because "
465 "it is not supported on the server\n");
466 clear_bit(LL_SBI_USER_XATTR, sbi->ll_flags);
469 if (data->ocd_connect_flags & OBD_CONNECT_ACL) {
471 sb->s_flags |= SB_POSIXACL;
473 set_bit(LL_SBI_ACL, sbi->ll_flags);
475 LCONSOLE_INFO("client wants to enable acl, but mdt not!\n");
477 sb->s_flags &= ~SB_POSIXACL;
479 clear_bit(LL_SBI_ACL, sbi->ll_flags);
482 if (data->ocd_connect_flags & OBD_CONNECT_64BITHASH)
483 set_bit(LL_SBI_64BIT_HASH, sbi->ll_flags);
485 if (data->ocd_connect_flags & OBD_CONNECT_LAYOUTLOCK)
486 set_bit(LL_SBI_LAYOUT_LOCK, sbi->ll_flags);
488 if (obd_connect_has_secctx(data))
489 set_bit(LL_SBI_FILE_SECCTX, sbi->ll_flags);
491 if (ll_sbi_has_encrypt(sbi) && !obd_connect_has_enc(data)) {
492 if (ll_sb_has_test_dummy_encryption(sb))
493 LCONSOLE_WARN("%s: server %s does not support encryption feature, encryption deactivated.\n",
495 sbi->ll_md_exp->exp_obd->obd_name);
496 ll_sbi_set_encrypt(sbi, false);
499 if (ll_sbi_has_name_encrypt(sbi) && !obd_connect_has_name_enc(data)) {
500 struct lustre_sb_info *lsi = s2lsi(sb);
502 if (ll_sb_has_test_dummy_encryption(sb))
503 LCONSOLE_WARN("%s: server %s does not support name encryption, not using it.\n",
505 sbi->ll_md_exp->exp_obd->obd_name);
506 lsi->lsi_flags &= ~LSI_FILENAME_ENC;
507 ll_sbi_set_name_encrypt(sbi, false);
510 if (data->ocd_ibits_known & MDS_INODELOCK_XATTR) {
511 if (!(data->ocd_connect_flags & OBD_CONNECT_MAX_EASIZE)) {
512 LCONSOLE_INFO("%s: disabling xattr cache due to "
513 "unknown maximum xattr size.\n", dt);
514 } else if (!sbi->ll_xattr_cache_set) {
515 /* If xattr_cache is already set (no matter 0 or 1)
516 * during processing llog, it won't be enabled here. */
517 set_bit(LL_SBI_XATTR_CACHE, sbi->ll_flags);
518 sbi->ll_xattr_cache_enabled = 1;
522 sbi->ll_dt_obd = class_name2obd(dt);
523 if (!sbi->ll_dt_obd) {
524 CERROR("DT %s: not setup or attached\n", dt);
525 GOTO(out_md_fid, err = -ENODEV);
528 /* pass client page size via ocd_grant_blkbits, the server should report
529 * back its backend blocksize for grant calculation purpose */
530 data->ocd_grant_blkbits = PAGE_SHIFT;
532 /* indicate OST features supported by this client */
533 data->ocd_connect_flags = OBD_CONNECT_GRANT | OBD_CONNECT_VERSION |
534 OBD_CONNECT_REQPORTAL | OBD_CONNECT_BRW_SIZE |
535 OBD_CONNECT_CANCELSET | OBD_CONNECT_FID |
536 OBD_CONNECT_SRVLOCK |
537 OBD_CONNECT_AT | OBD_CONNECT_OSS_CAPA |
538 OBD_CONNECT_VBR | OBD_CONNECT_FULL20 |
539 OBD_CONNECT_64BITHASH | OBD_CONNECT_MAXBYTES |
540 OBD_CONNECT_EINPROGRESS |
541 OBD_CONNECT_JOBSTATS | OBD_CONNECT_LVB_TYPE |
542 OBD_CONNECT_LAYOUTLOCK |
543 OBD_CONNECT_PINGLESS | OBD_CONNECT_LFSCK |
544 OBD_CONNECT_BULK_MBITS | OBD_CONNECT_SHORTIO |
545 OBD_CONNECT_FLAGS2 | OBD_CONNECT_GRANT_SHRINK;
546 data->ocd_connect_flags2 = OBD_CONNECT2_LOCKAHEAD |
547 OBD_CONNECT2_INC_XID | OBD_CONNECT2_LSEEK |
548 OBD_CONNECT2_REP_MBITS;
550 if (!OBD_FAIL_CHECK(OBD_FAIL_OSC_CONNECT_GRANT_PARAM))
551 data->ocd_connect_flags |= OBD_CONNECT_GRANT_PARAM;
553 /* OBD_CONNECT_CKSUM should always be set, even if checksums are
554 * disabled by default, because it can still be enabled on the
555 * fly via /sys. As a consequence, we still need to come to an
556 * agreement on the supported algorithms at connect time
558 data->ocd_connect_flags |= OBD_CONNECT_CKSUM;
560 if (OBD_FAIL_CHECK(OBD_FAIL_OSC_CKSUM_ADLER_ONLY))
561 data->ocd_cksum_types = OBD_CKSUM_ADLER;
563 data->ocd_cksum_types = obd_cksum_types_supported_client();
565 #ifdef HAVE_LRU_RESIZE_SUPPORT
566 data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
568 /* always ping even if server suppress_pings */
569 if (test_bit(LL_SBI_ALWAYS_PING, sbi->ll_flags))
570 data->ocd_connect_flags &= ~OBD_CONNECT_PINGLESS;
572 if (ll_sbi_has_encrypt(sbi))
573 obd_connect_set_enc(data);
575 CDEBUG(D_RPCTRACE, "ocd_connect_flags: %#llx ocd_version: %d "
576 "ocd_grant: %d\n", data->ocd_connect_flags,
577 data->ocd_version, data->ocd_grant);
579 sbi->ll_dt_obd->obd_upcall.onu_owner = &sbi->ll_lco;
580 sbi->ll_dt_obd->obd_upcall.onu_upcall = cl_ocd_update;
582 data->ocd_brw_size = DT_MAX_BRW_SIZE;
584 err = obd_connect(NULL, &sbi->ll_dt_exp, sbi->ll_dt_obd,
585 &sbi->ll_sb_uuid, data, sbi->ll_cache);
587 LCONSOLE_ERROR_MSG(0x150, "An OST (dt %s) is performing "
588 "recovery, of which this client is not a "
589 "part. Please wait for recovery to "
590 "complete, abort, or time out.\n", dt);
593 CERROR("%s: Cannot connect to %s: rc = %d\n",
594 sbi->ll_dt_exp->exp_obd->obd_name, dt, err);
598 if (ll_sbi_has_encrypt(sbi) &&
599 !obd_connect_has_enc(&sbi->ll_dt_obd->u.lov.lov_ocd)) {
600 if (ll_sb_has_test_dummy_encryption(sb))
601 LCONSOLE_WARN("%s: server %s does not support encryption feature, encryption deactivated.\n",
603 ll_sbi_set_encrypt(sbi, false);
604 } else if (ll_sb_has_test_dummy_encryption(sb)) {
605 LCONSOLE_WARN("Test dummy encryption mode enabled\n");
608 sbi->ll_dt_exp->exp_connect_data = *data;
610 /* Don't change value if it was specified in the config log */
611 if (sbi->ll_ra_info.ra_max_read_ahead_whole_pages == -1) {
612 sbi->ll_ra_info.ra_max_read_ahead_whole_pages =
613 max_t(unsigned long, SBI_DEFAULT_READ_AHEAD_WHOLE_MAX,
614 (data->ocd_brw_size >> PAGE_SHIFT));
615 if (sbi->ll_ra_info.ra_max_read_ahead_whole_pages >
616 sbi->ll_ra_info.ra_max_pages_per_file)
617 sbi->ll_ra_info.ra_max_read_ahead_whole_pages =
618 sbi->ll_ra_info.ra_max_pages_per_file;
621 err = obd_fid_init(sbi->ll_dt_exp->exp_obd, sbi->ll_dt_exp,
622 LUSTRE_SEQ_METADATA);
624 CERROR("%s: Can't init data layer FID infrastructure, "
625 "rc = %d\n", sbi->ll_dt_exp->exp_obd->obd_name, err);
629 mutex_lock(&sbi->ll_lco.lco_lock);
630 sbi->ll_lco.lco_flags = data->ocd_connect_flags;
631 sbi->ll_lco.lco_md_exp = sbi->ll_md_exp;
632 sbi->ll_lco.lco_dt_exp = sbi->ll_dt_exp;
633 mutex_unlock(&sbi->ll_lco.lco_lock);
635 fid_zero(&sbi->ll_root_fid);
636 err = md_get_root(sbi->ll_md_exp, get_mount_fileset(sb),
639 CERROR("cannot mds_connect: rc = %d\n", err);
640 GOTO(out_lock_cn_cb, err);
642 if (!fid_is_sane(&sbi->ll_root_fid)) {
643 CERROR("%s: Invalid root fid "DFID" during mount\n",
644 sbi->ll_md_exp->exp_obd->obd_name,
645 PFID(&sbi->ll_root_fid));
646 GOTO(out_lock_cn_cb, err = -EINVAL);
648 CDEBUG(D_SUPER, "rootfid "DFID"\n", PFID(&sbi->ll_root_fid));
650 sb->s_op = &lustre_super_operations;
651 sb->s_xattr = ll_xattr_handlers;
652 #if THREAD_SIZE >= 8192 /*b=17630*/
653 sb->s_export_op = &lustre_export_operations;
655 #ifdef HAVE_LUSTRE_CRYPTO
656 llcrypt_set_ops(sb, &lustre_cryptops);
660 * XXX: move this to after cbd setup? */
661 valid = OBD_MD_FLGETATTR | OBD_MD_FLBLOCKS | OBD_MD_FLMODEASIZE |
663 if (test_bit(LL_SBI_ACL, sbi->ll_flags))
664 valid |= OBD_MD_FLACL;
666 OBD_ALLOC_PTR(op_data);
668 GOTO(out_lock_cn_cb, err = -ENOMEM);
670 op_data->op_fid1 = sbi->ll_root_fid;
671 op_data->op_mode = 0;
672 op_data->op_valid = valid;
674 err = md_getattr(sbi->ll_md_exp, op_data, &request);
676 /* We need enc ctx info, so reset it in op_data to
677 * prevent it from being freed.
679 encctx = op_data->op_file_encctx;
680 encctxlen = op_data->op_file_encctx_size;
681 op_data->op_file_encctx = NULL;
682 op_data->op_file_encctx_size = 0;
683 OBD_FREE_PTR(op_data);
685 CERROR("%s: md_getattr failed for root: rc = %d\n",
686 sbi->ll_md_exp->exp_obd->obd_name, err);
687 GOTO(out_lock_cn_cb, err);
690 err = md_get_lustre_md(sbi->ll_md_exp, &request->rq_pill,
691 sbi->ll_dt_exp, sbi->ll_md_exp, &lmd);
693 CERROR("failed to understand root inode md: rc = %d\n", err);
694 ptlrpc_req_finished(request);
695 GOTO(out_lock_cn_cb, err);
698 LASSERT(fid_is_sane(&sbi->ll_root_fid));
699 api32 = test_bit(LL_SBI_32BIT_API, sbi->ll_flags);
700 root = ll_iget(sb, cl_fid_build_ino(&sbi->ll_root_fid, api32), &lmd);
701 md_free_lustre_md(sbi->ll_md_exp, &lmd);
705 err = IS_ERR(root) ? PTR_ERR(root) : -EBADF;
707 CERROR("%s: bad ll_iget() for root: rc = %d\n",
708 sbi->ll_fsname, err);
709 ptlrpc_req_finished(request);
715 "server returned encryption ctx for root inode "DFID"\n",
716 PFID(&sbi->ll_root_fid));
717 err = ll_set_encflags(root, encctx, encctxlen, true);
719 CWARN("%s: cannot set enc ctx for "DFID": rc = %d\n",
721 PFID(&sbi->ll_root_fid), err);
723 ptlrpc_req_finished(request);
725 checksum = test_bit(LL_SBI_CHECKSUM, sbi->ll_flags);
726 if (sbi->ll_checksum_set) {
727 err = obd_set_info_async(NULL, sbi->ll_dt_exp,
728 sizeof(KEY_CHECKSUM), KEY_CHECKSUM,
729 sizeof(checksum), &checksum, NULL);
731 CERROR("%s: Set checksum failed: rc = %d\n",
732 sbi->ll_dt_exp->exp_obd->obd_name, err);
738 sb->s_root = d_make_root(root);
739 if (sb->s_root == NULL) {
741 CERROR("%s: can't make root dentry: rc = %d\n",
742 sbi->ll_fsname, err);
746 sbi->ll_sdev_orig = sb->s_dev;
748 /* We set sb->s_dev equal on all lustre clients in order to support
749 * NFS export clustering. NFSD requires that the FSID be the same
751 /* s_dev is also used in lt_compare() to compare two fs, but that is
752 * only a node-local comparison. */
753 uuid = obd_get_uuid(sbi->ll_md_exp);
755 sb->s_dev = get_uuid2int(uuid->uuid, strlen(uuid->uuid));
762 if (sbi->ll_dt_obd) {
763 err = sysfs_create_link(&sbi->ll_kset.kobj,
764 &sbi->ll_dt_obd->obd_kset.kobj,
765 sbi->ll_dt_obd->obd_type->typ_name);
767 CERROR("%s: could not register %s in llite: rc = %d\n",
768 dt, sbi->ll_fsname, err);
773 if (sbi->ll_md_obd) {
774 err = sysfs_create_link(&sbi->ll_kset.kobj,
775 &sbi->ll_md_obd->obd_kset.kobj,
776 sbi->ll_md_obd->obd_type->typ_name);
778 CERROR("%s: could not register %s in llite: rc = %d\n",
779 md, sbi->ll_fsname, err);
788 obd_fid_fini(sbi->ll_dt_exp->exp_obd);
790 obd_disconnect(sbi->ll_dt_exp);
791 sbi->ll_dt_exp = NULL;
792 sbi->ll_dt_obd = NULL;
794 obd_fid_fini(sbi->ll_md_exp->exp_obd);
796 obd_disconnect(sbi->ll_md_exp);
797 sbi->ll_md_exp = NULL;
798 sbi->ll_md_obd = NULL;
807 int ll_get_max_mdsize(struct ll_sb_info *sbi, int *lmmsize)
811 size = sizeof(*lmmsize);
812 rc = obd_get_info(NULL, sbi->ll_dt_exp, sizeof(KEY_MAX_EASIZE),
813 KEY_MAX_EASIZE, &size, lmmsize);
815 CERROR("%s: cannot get max LOV EA size: rc = %d\n",
816 sbi->ll_dt_exp->exp_obd->obd_name, rc);
820 CDEBUG(D_INFO, "max LOV ea size: %d\n", *lmmsize);
823 rc = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_MAX_EASIZE),
824 KEY_MAX_EASIZE, &size, lmmsize);
826 CERROR("Get max mdsize error rc %d\n", rc);
828 CDEBUG(D_INFO, "max LMV ea size: %d\n", *lmmsize);
834 * Get the value of the default_easize parameter.
836 * \see client_obd::cl_default_mds_easize
838 * \param[in] sbi superblock info for this filesystem
839 * \param[out] lmmsize pointer to storage location for value
841 * \retval 0 on success
842 * \retval negative negated errno on failure
844 int ll_get_default_mdsize(struct ll_sb_info *sbi, int *lmmsize)
849 rc = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_DEFAULT_EASIZE),
850 KEY_DEFAULT_EASIZE, &size, lmmsize);
852 CERROR("Get default mdsize error rc %d\n", rc);
858 * Set the default_easize parameter to the given value.
860 * \see client_obd::cl_default_mds_easize
862 * \param[in] sbi superblock info for this filesystem
863 * \param[in] lmmsize the size to set
865 * \retval 0 on success
866 * \retval negative negated errno on failure
868 int ll_set_default_mdsize(struct ll_sb_info *sbi, int lmmsize)
872 if (lmmsize < sizeof(struct lov_mds_md) ||
873 lmmsize > OBD_MAX_DEFAULT_EA_SIZE)
876 rc = obd_set_info_async(NULL, sbi->ll_md_exp,
877 sizeof(KEY_DEFAULT_EASIZE), KEY_DEFAULT_EASIZE,
878 sizeof(int), &lmmsize, NULL);
883 static void client_common_put_super(struct super_block *sb)
885 struct ll_sb_info *sbi = ll_s2sbi(sb);
890 obd_fid_fini(sbi->ll_dt_exp->exp_obd);
891 obd_disconnect(sbi->ll_dt_exp);
892 sbi->ll_dt_exp = NULL;
894 ll_debugfs_unregister_super(sb);
896 obd_fid_fini(sbi->ll_md_exp->exp_obd);
897 obd_disconnect(sbi->ll_md_exp);
898 sbi->ll_md_exp = NULL;
903 void ll_kill_super(struct super_block *sb)
905 struct ll_sb_info *sbi;
909 if (!(sb->s_flags & SB_ACTIVE))
913 /* we need restore s_dev from changed for clustred NFS before put_super
914 * because new kernels have cached s_dev and change sb->s_dev in
915 * put_super not affected real removing devices */
917 sb->s_dev = sbi->ll_sdev_orig;
919 /* wait running statahead threads to quit */
920 while (atomic_read(&sbi->ll_sa_running) > 0)
921 schedule_timeout_uninterruptible(
922 cfs_time_seconds(1) >> 3);
928 /* Since we use this table for ll_sbi_flags_seq_show make
929 * sure what you want displayed for a specific token that
930 * is listed more than once below be listed first. For
931 * example we want "checksum" displayed, not "nochecksum"
934 static const match_table_t ll_sbi_flags_name = {
935 {LL_SBI_NOLCK, "nolock"},
936 {LL_SBI_CHECKSUM, "checksum"},
937 {LL_SBI_CHECKSUM, "nochecksum"},
938 {LL_SBI_LOCALFLOCK, "localflock"},
939 {LL_SBI_FLOCK, "flock"},
940 {LL_SBI_FLOCK, "noflock"},
941 {LL_SBI_USER_XATTR, "user_xattr"},
942 {LL_SBI_USER_XATTR, "nouser_xattr"},
943 {LL_SBI_LRU_RESIZE, "lruresize"},
944 {LL_SBI_LRU_RESIZE, "nolruresize"},
945 {LL_SBI_LAZYSTATFS, "lazystatfs"},
946 {LL_SBI_LAZYSTATFS, "nolazystatfs"},
947 {LL_SBI_32BIT_API, "32bitapi"},
948 {LL_SBI_USER_FID2PATH, "user_fid2path"},
949 {LL_SBI_USER_FID2PATH, "nouser_fid2path"},
950 {LL_SBI_VERBOSE, "verbose"},
951 {LL_SBI_VERBOSE, "noverbose"},
952 {LL_SBI_ALWAYS_PING, "always_ping"},
953 {LL_SBI_TEST_DUMMY_ENCRYPTION, "test_dummy_encryption=%s"},
954 {LL_SBI_TEST_DUMMY_ENCRYPTION, "test_dummy_encryption"},
955 {LL_SBI_ENCRYPT, "encrypt"},
956 {LL_SBI_ENCRYPT, "noencrypt"},
957 {LL_SBI_FOREIGN_SYMLINK, "foreign_symlink=%s"},
958 {LL_SBI_NUM_MOUNT_OPT, NULL},
961 {LL_SBI_AGL_ENABLED, "agl"},
962 {LL_SBI_64BIT_HASH, "64bit_hash"},
963 {LL_SBI_LAYOUT_LOCK, "layout"},
964 {LL_SBI_XATTR_CACHE, "xattr_cache"},
965 {LL_SBI_NOROOTSQUASH, "norootsquash"},
966 {LL_SBI_FAST_READ, "fast_read"},
967 {LL_SBI_FILE_SECCTX, "file_secctx"},
968 {LL_SBI_TINY_WRITE, "tiny_write"},
969 {LL_SBI_FILE_HEAT, "file_heat"},
970 {LL_SBI_PARALLEL_DIO, "parallel_dio"},
971 {LL_SBI_ENCRYPT_NAME, "name_encrypt"},
974 int ll_sbi_flags_seq_show(struct seq_file *m, void *v)
976 struct super_block *sb = m->private;
979 for (i = 0; i < LL_SBI_NUM_FLAGS; i++) {
982 if (!test_bit(i, ll_s2sbi(sb)->ll_flags))
985 for (j = 0; j < ARRAY_SIZE(ll_sbi_flags_name); j++) {
986 if (ll_sbi_flags_name[j].token == i &&
987 ll_sbi_flags_name[j].pattern) {
989 ll_sbi_flags_name[j].pattern);
998 /* non-client-specific mount options are parsed in lmd_parse */
999 static int ll_options(char *options, struct super_block *sb)
1001 struct ll_sb_info *sbi = ll_s2sbi(sb);
1002 char *s2, *s1, *opts;
1009 /* Don't stomp on lmd_opts */
1010 opts = kstrdup(options, GFP_KERNEL);
1016 CDEBUG(D_CONFIG, "Parsing opts %s\n", options);
1018 while ((s1 = strsep(&opts, ",")) != NULL) {
1019 substring_t args[MAX_OPT_ARGS];
1020 bool turn_off = false;
1026 CDEBUG(D_SUPER, "next opt=%s\n", s1);
1028 if (strncmp(s1, "no", 2) == 0)
1032 * Initialize args struct so we know whether arg was
1033 * found; some options take optional arguments.
1036 args[0].from = NULL;
1037 token = match_token(s1, ll_sbi_flags_name, args);
1038 if (token == LL_SBI_NUM_MOUNT_OPT) {
1039 if (match_wildcard("context", s1) ||
1040 match_wildcard("fscontext", s1) ||
1041 match_wildcard("defcontext", s1) ||
1042 match_wildcard("rootcontext",s1))
1045 LCONSOLE_ERROR_MSG(0x152,
1046 "Unknown option '%s', won't mount.\n",
1053 case LL_SBI_32BIT_API:
1054 case LL_SBI_64BIT_HASH:
1055 case LL_SBI_ALWAYS_PING:
1056 set_bit(token, sbi->ll_flags);
1060 clear_bit(LL_SBI_LOCALFLOCK, sbi->ll_flags);
1062 clear_bit(LL_SBI_FLOCK, sbi->ll_flags);
1064 set_bit(token, sbi->ll_flags);
1067 case LL_SBI_LOCALFLOCK:
1068 clear_bit(LL_SBI_FLOCK, sbi->ll_flags);
1069 set_bit(token, sbi->ll_flags);
1072 case LL_SBI_CHECKSUM:
1073 sbi->ll_checksum_set = 1;
1075 case LL_SBI_USER_XATTR:
1076 case LL_SBI_USER_FID2PATH:
1077 case LL_SBI_LRU_RESIZE:
1078 case LL_SBI_LAZYSTATFS:
1079 case LL_SBI_VERBOSE:
1081 clear_bit(token, sbi->ll_flags);
1083 set_bit(token, sbi->ll_flags);
1085 case LL_SBI_TEST_DUMMY_ENCRYPTION: {
1086 #ifdef HAVE_LUSTRE_CRYPTO
1087 #ifdef HAVE_FSCRYPT_DUMMY_CONTEXT_ENABLED
1088 set_bit(token, sbi->ll_flags);
1090 struct lustre_sb_info *lsi = s2lsi(sb);
1092 err = llcrypt_set_test_dummy_encryption(sb, &args[0],
1093 &lsi->lsi_dummy_enc_ctx);
1099 "Can't change test_dummy_encryption");
1100 else if (err == -EINVAL)
1102 "Value of option \"%s\" unrecognized",
1106 "Error processing option \"%s\" [%d]",
1111 LCONSOLE_WARN("Test dummy encryption mount option ignored: encryption not supported\n");
1115 case LL_SBI_ENCRYPT:
1116 #ifdef HAVE_LUSTRE_CRYPTO
1118 clear_bit(token, sbi->ll_flags);
1120 set_bit(token, sbi->ll_flags);
1122 LCONSOLE_WARN("noencrypt or encrypt mount option ignored: encryption not supported\n");
1125 case LL_SBI_FOREIGN_SYMLINK:
1126 /* non-default prefix provided ? */
1131 /* path must be absolute */
1132 if (args->from[0] != '/') {
1133 LCONSOLE_ERROR_MSG(0x152,
1134 "foreign prefix '%s' must be an absolute path\n",
1139 old_len = sbi->ll_foreign_symlink_prefix_size;
1140 old = sbi->ll_foreign_symlink_prefix;
1141 /* alloc for path length and '\0' */
1142 sbi->ll_foreign_symlink_prefix = match_strdup(args);
1143 if (!sbi->ll_foreign_symlink_prefix) {
1144 /* restore previous */
1145 sbi->ll_foreign_symlink_prefix = old;
1146 sbi->ll_foreign_symlink_prefix_size =
1150 sbi->ll_foreign_symlink_prefix_size =
1151 args->to - args->from + 1;
1152 OBD_ALLOC_POST(sbi->ll_foreign_symlink_prefix,
1153 sbi->ll_foreign_symlink_prefix_size,
1156 OBD_FREE(old, old_len);
1158 /* enable foreign symlink support */
1159 set_bit(token, sbi->ll_flags);
1161 LCONSOLE_ERROR_MSG(0x152,
1162 "invalid %s option\n", s1);
1173 void ll_lli_init(struct ll_inode_info *lli)
1175 lli->lli_inode_magic = LLI_INODE_MAGIC;
1177 rwlock_init(&lli->lli_lock);
1178 lli->lli_posix_acl = NULL;
1179 /* Do not set lli_fid, it has been initialized already. */
1180 fid_zero(&lli->lli_pfid);
1181 lli->lli_mds_read_och = NULL;
1182 lli->lli_mds_write_och = NULL;
1183 lli->lli_mds_exec_och = NULL;
1184 lli->lli_open_fd_read_count = 0;
1185 lli->lli_open_fd_write_count = 0;
1186 lli->lli_open_fd_exec_count = 0;
1187 mutex_init(&lli->lli_och_mutex);
1188 spin_lock_init(&lli->lli_agl_lock);
1189 spin_lock_init(&lli->lli_layout_lock);
1190 ll_layout_version_set(lli, CL_LAYOUT_GEN_NONE);
1191 lli->lli_clob = NULL;
1193 init_rwsem(&lli->lli_xattrs_list_rwsem);
1194 mutex_init(&lli->lli_xattrs_enq_lock);
1196 LASSERT(lli->lli_vfs_inode.i_mode != 0);
1197 if (S_ISDIR(lli->lli_vfs_inode.i_mode)) {
1198 lli->lli_opendir_key = NULL;
1199 lli->lli_sai = NULL;
1200 spin_lock_init(&lli->lli_sa_lock);
1201 lli->lli_opendir_pid = 0;
1202 lli->lli_sa_enabled = 0;
1203 init_rwsem(&lli->lli_lsm_sem);
1205 mutex_init(&lli->lli_size_mutex);
1206 mutex_init(&lli->lli_setattr_mutex);
1207 lli->lli_symlink_name = NULL;
1208 ll_trunc_sem_init(&lli->lli_trunc_sem);
1209 range_lock_tree_init(&lli->lli_write_tree);
1210 init_rwsem(&lli->lli_glimpse_sem);
1211 lli->lli_glimpse_time = ktime_set(0, 0);
1212 INIT_LIST_HEAD(&lli->lli_agl_list);
1213 lli->lli_agl_index = 0;
1214 lli->lli_async_rc = 0;
1215 spin_lock_init(&lli->lli_heat_lock);
1216 obd_heat_clear(lli->lli_heat_instances, OBD_HEAT_COUNT);
1217 lli->lli_heat_flags = 0;
1218 mutex_init(&lli->lli_pcc_lock);
1219 lli->lli_pcc_state = PCC_STATE_FL_NONE;
1220 lli->lli_pcc_inode = NULL;
1221 lli->lli_pcc_dsflags = PCC_DATASET_INVALID;
1222 lli->lli_pcc_generation = 0;
1223 mutex_init(&lli->lli_group_mutex);
1224 lli->lli_group_users = 0;
1225 lli->lli_group_gid = 0;
1227 mutex_init(&lli->lli_layout_mutex);
1228 memset(lli->lli_jobid, 0, sizeof(lli->lli_jobid));
1229 /* ll_cl_context initialize */
1230 INIT_LIST_HEAD(&lli->lli_lccs);
1233 #define MAX_STRING_SIZE 128
1235 #ifndef HAVE_SUPER_SETUP_BDI_NAME
1236 #ifndef HAVE_BDI_CAP_MAP_COPY
1237 # define BDI_CAP_MAP_COPY 0
1240 static int super_setup_bdi_name(struct super_block *sb, char *fmt, ...)
1242 struct lustre_sb_info *lsi = s2lsi(sb);
1243 char buf[MAX_STRING_SIZE];
1247 err = bdi_init(&lsi->lsi_bdi);
1251 lsi->lsi_flags |= LSI_BDI_INITIALIZED;
1252 lsi->lsi_bdi.capabilities = BDI_CAP_MAP_COPY;
1253 lsi->lsi_bdi.name = "lustre";
1254 va_start(args, fmt);
1255 vsnprintf(buf, MAX_STRING_SIZE, fmt, args);
1257 err = bdi_register(&lsi->lsi_bdi, NULL, "%s", buf);
1260 sb->s_bdi = &lsi->lsi_bdi;
1264 #endif /* !HAVE_SUPER_SETUP_BDI_NAME */
1266 int ll_fill_super(struct super_block *sb)
1268 struct lustre_profile *lprof = NULL;
1269 struct lustre_sb_info *lsi = s2lsi(sb);
1270 struct ll_sb_info *sbi = NULL;
1271 char *dt = NULL, *md = NULL;
1272 char *profilenm = get_profile_name(sb);
1273 struct config_llog_instance *cfg;
1274 /* %p for void* in printf needs 16+2 characters: 0xffffffffffffffff */
1275 const int instlen = LUSTRE_MAXINSTANCE + 2;
1276 unsigned long cfg_instance = ll_get_cfg_instance(sb);
1277 char name[MAX_STRING_SIZE];
1286 /* for ASLR, to map between cfg_instance and hashed ptr */
1287 CDEBUG(D_VFSTRACE, "VFS Op: cfg_instance %s-%016lx (sb %p)\n",
1288 profilenm, cfg_instance, sb);
1290 OBD_RACE(OBD_FAIL_LLITE_RACE_MOUNT);
1294 GOTO(out_free_cfg, err = -ENOMEM);
1296 /* client additional sb info */
1297 lsi->lsi_llsbi = sbi = ll_init_sbi();
1299 GOTO(out_free_cfg, err = PTR_ERR(sbi));
1301 err = ll_options(lsi->lsi_lmd->lmd_opts, sb);
1303 GOTO(out_free_cfg, err);
1305 /* LSI_FILENAME_ENC is only used by embedded llcrypt */
1306 #ifdef CONFIG_LL_ENCRYPTION
1307 if (ll_sb_has_test_dummy_encryption(sb))
1308 /* enable filename encryption by default for dummy enc mode */
1309 lsi->lsi_flags |= LSI_FILENAME_ENC;
1311 /* filename encryption is disabled by default */
1312 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;
1354 /* Call ll_debugfs_register_super() before lustre_process_log()
1355 * so that "llite.*.*" params can be processed correctly.
1357 err = ll_debugfs_register_super(sb, name);
1359 CERROR("%s: could not register mountpoint in llite: rc = %d\n",
1360 sbi->ll_fsname, err);
1364 /* The cfg_instance is a value unique to this super, in case some
1365 * joker tries to mount the same fs at two mount points.
1367 cfg->cfg_instance = cfg_instance;
1368 cfg->cfg_uuid = lsi->lsi_llsbi->ll_sb_uuid;
1369 cfg->cfg_callback = class_config_llog_handler;
1370 cfg->cfg_sub_clds = CONFIG_SUB_CLIENT;
1371 /* set up client obds */
1372 err = lustre_process_log(sb, profilenm, cfg);
1374 GOTO(out_debugfs, err);
1376 /* Profile set with LCFG_MOUNTOPT so we can find our mdc and osc obds */
1377 lprof = class_get_profile(profilenm);
1378 if (lprof == NULL) {
1379 LCONSOLE_ERROR_MSG(0x156, "The client profile '%s' could not be"
1380 " read from the MGS. Does that filesystem "
1381 "exist?\n", profilenm);
1382 GOTO(out_debugfs, err = -EINVAL);
1384 CDEBUG(D_CONFIG, "Found profile %s: mdc=%s osc=%s\n", profilenm,
1385 lprof->lp_md, lprof->lp_dt);
1387 dt_len = strlen(lprof->lp_dt) + instlen + 2;
1388 OBD_ALLOC(dt, dt_len);
1390 GOTO(out_profile, err = -ENOMEM);
1391 snprintf(dt, dt_len - 1, "%s-%016lx", lprof->lp_dt, cfg_instance);
1393 md_len = strlen(lprof->lp_md) + instlen + 2;
1394 OBD_ALLOC(md, md_len);
1396 GOTO(out_free_dt, err = -ENOMEM);
1397 snprintf(md, md_len - 1, "%s-%016lx", lprof->lp_md, cfg_instance);
1399 /* connections, registrations, sb setup */
1400 err = client_common_fill_super(sb, md, dt);
1402 GOTO(out_free_md, err);
1404 sbi->ll_client_common_fill_super_succeeded = 1;
1408 OBD_FREE(md, md_len);
1411 OBD_FREE(dt, dt_len);
1414 class_put_profile(lprof);
1417 ll_debugfs_unregister_super(sb);
1424 else if (test_bit(LL_SBI_VERBOSE, sbi->ll_flags))
1425 LCONSOLE_WARN("Mounted %s\n", profilenm);
1427 } /* ll_fill_super */
1429 void ll_put_super(struct super_block *sb)
1431 struct config_llog_instance cfg, params_cfg;
1432 struct obd_device *obd;
1433 struct lustre_sb_info *lsi = s2lsi(sb);
1434 struct ll_sb_info *sbi = ll_s2sbi(sb);
1435 char *profilenm = get_profile_name(sb);
1436 unsigned long cfg_instance = ll_get_cfg_instance(sb);
1438 int next, force = 1, rc = 0;
1442 GOTO(out_no_sbi, 0);
1444 /* Should replace instance_id with something better for ASLR */
1445 CDEBUG(D_VFSTRACE, "VFS Op: cfg_instance %s-%016lx (sb %p)\n",
1446 profilenm, cfg_instance, sb);
1448 cfg.cfg_instance = cfg_instance;
1449 lustre_end_log(sb, profilenm, &cfg);
1451 params_cfg.cfg_instance = cfg_instance;
1452 lustre_end_log(sb, PARAMS_FILENAME, ¶ms_cfg);
1454 if (sbi->ll_md_exp) {
1455 obd = class_exp2obd(sbi->ll_md_exp);
1457 force = obd->obd_force;
1460 /* Wait for unstable pages to be committed to stable storage */
1462 rc = l_wait_event_abortable(
1463 sbi->ll_cache->ccc_unstable_waitq,
1464 atomic_long_read(&sbi->ll_cache->ccc_unstable_nr) == 0);
1467 ccc_count = atomic_long_read(&sbi->ll_cache->ccc_unstable_nr);
1468 if (force == 0 && rc != -ERESTARTSYS)
1469 LASSERTF(ccc_count == 0, "count: %li\n", ccc_count);
1471 /* We need to set force before the lov_disconnect in
1472 * lustre_common_put_super, since l_d cleans up osc's as well.
1476 while ((obd = class_devices_in_group(&sbi->ll_sb_uuid,
1478 obd->obd_force = force;
1482 if (sbi->ll_client_common_fill_super_succeeded) {
1483 /* Only if client_common_fill_super succeeded */
1484 client_common_put_super(sb);
1488 while ((obd = class_devices_in_group(&sbi->ll_sb_uuid, &next)))
1489 class_manual_cleanup(obd);
1491 if (test_bit(LL_SBI_VERBOSE, sbi->ll_flags))
1492 LCONSOLE_WARN("Unmounted %s\n", profilenm ? profilenm : "");
1495 class_del_profile(profilenm);
1497 #ifndef HAVE_SUPER_SETUP_BDI_NAME
1498 if (lsi->lsi_flags & LSI_BDI_INITIALIZED) {
1499 bdi_destroy(&lsi->lsi_bdi);
1500 lsi->lsi_flags &= ~LSI_BDI_INITIALIZED;
1504 llcrypt_free_dummy_context(&lsi->lsi_dummy_enc_ctx);
1506 lsi->lsi_llsbi = NULL;
1508 lustre_common_put_super(sb);
1510 cl_env_cache_purge(~0);
1513 } /* client_put_super */
1515 struct inode *ll_inode_from_resource_lock(struct ldlm_lock *lock)
1517 struct inode *inode = NULL;
1519 /* NOTE: we depend on atomic igrab() -bzzz */
1520 lock_res_and_lock(lock);
1521 if (lock->l_resource->lr_lvb_inode) {
1522 struct ll_inode_info * lli;
1523 lli = ll_i2info(lock->l_resource->lr_lvb_inode);
1524 if (lli->lli_inode_magic == LLI_INODE_MAGIC) {
1525 inode = igrab(lock->l_resource->lr_lvb_inode);
1527 inode = lock->l_resource->lr_lvb_inode;
1528 LDLM_DEBUG_LIMIT(inode->i_state & I_FREEING ? D_INFO :
1529 D_WARNING, lock, "lr_lvb_inode %p is "
1530 "bogus: magic %08x",
1531 lock->l_resource->lr_lvb_inode,
1532 lli->lli_inode_magic);
1536 unlock_res_and_lock(lock);
1540 void ll_dir_clear_lsm_md(struct inode *inode)
1542 struct ll_inode_info *lli = ll_i2info(inode);
1544 LASSERT(S_ISDIR(inode->i_mode));
1546 if (lli->lli_lsm_md) {
1547 lmv_free_memmd(lli->lli_lsm_md);
1548 lli->lli_lsm_md = NULL;
1551 if (lli->lli_default_lsm_md) {
1552 lmv_free_memmd(lli->lli_default_lsm_md);
1553 lli->lli_default_lsm_md = NULL;
1557 static struct inode *ll_iget_anon_dir(struct super_block *sb,
1558 const struct lu_fid *fid,
1559 struct lustre_md *md)
1561 struct ll_sb_info *sbi = ll_s2sbi(sb);
1562 struct ll_inode_info *lli;
1563 struct mdt_body *body = md->body;
1564 struct inode *inode;
1570 ino = cl_fid_build_ino(fid, test_bit(LL_SBI_32BIT_API, sbi->ll_flags));
1571 inode = iget_locked(sb, ino);
1572 if (inode == NULL) {
1573 CERROR("%s: failed get simple inode "DFID": rc = -ENOENT\n",
1574 sbi->ll_fsname, PFID(fid));
1575 RETURN(ERR_PTR(-ENOENT));
1578 lli = ll_i2info(inode);
1579 if (inode->i_state & I_NEW) {
1580 inode->i_mode = (inode->i_mode & ~S_IFMT) |
1581 (body->mbo_mode & S_IFMT);
1582 LASSERTF(S_ISDIR(inode->i_mode), "Not slave inode "DFID"\n",
1585 inode->i_mtime.tv_sec = 0;
1586 inode->i_atime.tv_sec = 0;
1587 inode->i_ctime.tv_sec = 0;
1590 #ifdef HAVE_BACKING_DEV_INFO
1591 /* initializing backing dev info. */
1592 inode->i_mapping->backing_dev_info =
1593 &s2lsi(inode->i_sb)->lsi_bdi;
1595 inode->i_op = &ll_dir_inode_operations;
1596 inode->i_fop = &ll_dir_operations;
1597 lli->lli_fid = *fid;
1600 /* master object FID */
1601 lli->lli_pfid = body->mbo_fid1;
1602 CDEBUG(D_INODE, "lli %p slave "DFID" master "DFID"\n",
1603 lli, PFID(fid), PFID(&lli->lli_pfid));
1604 unlock_new_inode(inode);
1606 /* in directory restripe/auto-split, a directory will be
1607 * transformed to a stripe if it's plain, set its pfid here,
1608 * otherwise ll_lock_cancel_bits() can't find the master inode.
1610 lli->lli_pfid = body->mbo_fid1;
1616 static int ll_init_lsm_md(struct inode *inode, struct lustre_md *md)
1619 struct lmv_stripe_md *lsm = md->lmv;
1620 struct ll_inode_info *lli = ll_i2info(inode);
1623 LASSERT(lsm != NULL);
1625 CDEBUG(D_INODE, "%s: "DFID" set dir layout:\n",
1626 ll_i2sbi(inode)->ll_fsname, PFID(&lli->lli_fid));
1627 lsm_md_dump(D_INODE, lsm);
1629 if (!lmv_dir_striped(lsm))
1632 /* XXX sigh, this lsm_root initialization should be in
1633 * LMV layer, but it needs ll_iget right now, so we
1634 * put this here right now. */
1635 for (i = 0; i < lsm->lsm_md_stripe_count; i++) {
1636 fid = &lsm->lsm_md_oinfo[i].lmo_fid;
1637 LASSERT(lsm->lsm_md_oinfo[i].lmo_root == NULL);
1639 if (!fid_is_sane(fid))
1642 /* Unfortunately ll_iget will call ll_update_inode,
1643 * where the initialization of slave inode is slightly
1644 * different, so it reset lsm_md to NULL to avoid
1645 * initializing lsm for slave inode. */
1646 lsm->lsm_md_oinfo[i].lmo_root =
1647 ll_iget_anon_dir(inode->i_sb, fid, md);
1648 if (IS_ERR(lsm->lsm_md_oinfo[i].lmo_root)) {
1649 int rc = PTR_ERR(lsm->lsm_md_oinfo[i].lmo_root);
1651 lsm->lsm_md_oinfo[i].lmo_root = NULL;
1653 iput(lsm->lsm_md_oinfo[i].lmo_root);
1654 lsm->lsm_md_oinfo[i].lmo_root = NULL;
1660 lli->lli_lsm_md = lsm;
1665 static void ll_update_default_lsm_md(struct inode *inode, struct lustre_md *md)
1667 struct ll_inode_info *lli = ll_i2info(inode);
1671 if (!md->default_lmv) {
1672 /* clear default lsm */
1673 if (lli->lli_default_lsm_md) {
1674 down_write(&lli->lli_lsm_sem);
1675 if (lli->lli_default_lsm_md) {
1676 lmv_free_memmd(lli->lli_default_lsm_md);
1677 lli->lli_default_lsm_md = NULL;
1679 up_write(&lli->lli_lsm_sem);
1684 if (lli->lli_default_lsm_md) {
1685 /* do nonthing if default lsm isn't changed */
1686 down_read(&lli->lli_lsm_sem);
1687 if (lli->lli_default_lsm_md &&
1688 lsm_md_eq(lli->lli_default_lsm_md, md->default_lmv)) {
1689 up_read(&lli->lli_lsm_sem);
1692 up_read(&lli->lli_lsm_sem);
1695 down_write(&lli->lli_lsm_sem);
1696 if (lli->lli_default_lsm_md)
1697 lmv_free_memmd(lli->lli_default_lsm_md);
1698 lli->lli_default_lsm_md = md->default_lmv;
1699 lsm_md_dump(D_INODE, md->default_lmv);
1700 md->default_lmv = NULL;
1701 up_write(&lli->lli_lsm_sem);
1705 static int ll_update_lsm_md(struct inode *inode, struct lustre_md *md)
1707 struct ll_inode_info *lli = ll_i2info(inode);
1708 struct lmv_stripe_md *lsm = md->lmv;
1709 struct cl_attr *attr;
1714 LASSERT(S_ISDIR(inode->i_mode));
1715 CDEBUG(D_INODE, "update lsm %p of "DFID"\n", lli->lli_lsm_md,
1716 PFID(ll_inode2fid(inode)));
1718 /* update default LMV */
1719 if (md->default_lmv)
1720 ll_update_default_lsm_md(inode, md);
1722 /* after dir migration/restripe, a stripe may be turned into a
1723 * directory, in this case, zero out its lli_pfid.
1725 if (unlikely(fid_is_norm(&lli->lli_pfid)))
1726 fid_zero(&lli->lli_pfid);
1729 * no striped information from request, lustre_md from req does not
1730 * include stripeEA, see ll_md_setattr()
1736 * normally dir layout doesn't change, only take read lock to check
1737 * that to avoid blocking other MD operations.
1739 down_read(&lli->lli_lsm_sem);
1741 /* some current lookup initialized lsm, and unchanged */
1742 if (lli->lli_lsm_md && lsm_md_eq(lli->lli_lsm_md, lsm))
1743 GOTO(unlock, rc = 0);
1745 /* if dir layout doesn't match, check whether version is increased,
1746 * which means layout is changed, this happens in dir split/merge and
1749 * foreign LMV should not change.
1751 if (lli->lli_lsm_md && lmv_dir_striped(lli->lli_lsm_md) &&
1752 lsm->lsm_md_layout_version <=
1753 lli->lli_lsm_md->lsm_md_layout_version) {
1754 CERROR("%s: "DFID" dir layout mismatch:\n",
1755 ll_i2sbi(inode)->ll_fsname, PFID(&lli->lli_fid));
1756 lsm_md_dump(D_ERROR, lli->lli_lsm_md);
1757 lsm_md_dump(D_ERROR, lsm);
1758 GOTO(unlock, rc = -EINVAL);
1761 up_read(&lli->lli_lsm_sem);
1762 down_write(&lli->lli_lsm_sem);
1763 /* clear existing lsm */
1764 if (lli->lli_lsm_md) {
1765 lmv_free_memmd(lli->lli_lsm_md);
1766 lli->lli_lsm_md = NULL;
1769 rc = ll_init_lsm_md(inode, md);
1770 up_write(&lli->lli_lsm_sem);
1775 /* set md->lmv to NULL, so the following free lustre_md will not free
1780 /* md_merge_attr() may take long, since lsm is already set, switch to
1783 down_read(&lli->lli_lsm_sem);
1785 if (!lmv_dir_striped(lli->lli_lsm_md))
1786 GOTO(unlock, rc = 0);
1788 OBD_ALLOC_PTR(attr);
1790 GOTO(unlock, rc = -ENOMEM);
1792 /* validate the lsm */
1793 rc = md_merge_attr(ll_i2mdexp(inode), lli->lli_lsm_md, attr,
1794 ll_md_blocking_ast);
1796 if (md->body->mbo_valid & OBD_MD_FLNLINK)
1797 md->body->mbo_nlink = attr->cat_nlink;
1798 if (md->body->mbo_valid & OBD_MD_FLSIZE)
1799 md->body->mbo_size = attr->cat_size;
1800 if (md->body->mbo_valid & OBD_MD_FLATIME)
1801 md->body->mbo_atime = attr->cat_atime;
1802 if (md->body->mbo_valid & OBD_MD_FLCTIME)
1803 md->body->mbo_ctime = attr->cat_ctime;
1804 if (md->body->mbo_valid & OBD_MD_FLMTIME)
1805 md->body->mbo_mtime = attr->cat_mtime;
1811 up_read(&lli->lli_lsm_sem);
1816 void ll_clear_inode(struct inode *inode)
1818 struct ll_inode_info *lli = ll_i2info(inode);
1819 struct ll_sb_info *sbi = ll_i2sbi(inode);
1823 CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p)\n",
1824 PFID(ll_inode2fid(inode)), inode);
1826 if (S_ISDIR(inode->i_mode)) {
1827 /* these should have been cleared in ll_file_release */
1828 LASSERT(lli->lli_opendir_key == NULL);
1829 LASSERT(lli->lli_sai == NULL);
1830 LASSERT(lli->lli_opendir_pid == 0);
1832 pcc_inode_free(inode);
1835 md_null_inode(sbi->ll_md_exp, ll_inode2fid(inode));
1837 LASSERT(!lli->lli_open_fd_write_count);
1838 LASSERT(!lli->lli_open_fd_read_count);
1839 LASSERT(!lli->lli_open_fd_exec_count);
1841 if (lli->lli_mds_write_och)
1842 ll_md_real_close(inode, FMODE_WRITE);
1843 if (lli->lli_mds_exec_och)
1844 ll_md_real_close(inode, FMODE_EXEC);
1845 if (lli->lli_mds_read_och)
1846 ll_md_real_close(inode, FMODE_READ);
1848 if (S_ISLNK(inode->i_mode) && lli->lli_symlink_name) {
1849 OBD_FREE(lli->lli_symlink_name,
1850 strlen(lli->lli_symlink_name) + 1);
1851 lli->lli_symlink_name = NULL;
1854 ll_xattr_cache_destroy(inode);
1856 forget_all_cached_acls(inode);
1858 lli->lli_inode_magic = LLI_INODE_DEAD;
1860 if (S_ISDIR(inode->i_mode))
1861 ll_dir_clear_lsm_md(inode);
1862 else if (S_ISREG(inode->i_mode) && !is_bad_inode(inode))
1863 LASSERT(list_empty(&lli->lli_agl_list));
1866 * XXX This has to be done before lsm is freed below, because
1867 * cl_object still uses inode lsm.
1869 cl_inode_fini(inode);
1871 llcrypt_put_encryption_info(inode);
1876 static int ll_md_setattr(struct dentry *dentry, struct md_op_data *op_data)
1878 struct lustre_md md;
1879 struct inode *inode = dentry->d_inode;
1880 struct ll_sb_info *sbi = ll_i2sbi(inode);
1881 struct ptlrpc_request *request = NULL;
1886 op_data = ll_prep_md_op_data(op_data, inode, NULL, NULL, 0, 0,
1887 LUSTRE_OPC_ANY, NULL);
1888 if (IS_ERR(op_data))
1889 RETURN(PTR_ERR(op_data));
1891 /* If this is a chgrp of a regular file, we want to reserve enough
1892 * quota to cover the entire file size.
1894 if (S_ISREG(inode->i_mode) && op_data->op_attr.ia_valid & ATTR_GID &&
1895 from_kgid(&init_user_ns, op_data->op_attr.ia_gid) !=
1896 from_kgid(&init_user_ns, inode->i_gid)) {
1897 op_data->op_xvalid |= OP_XVALID_BLOCKS;
1898 op_data->op_attr_blocks = inode->i_blocks;
1902 rc = md_setattr(sbi->ll_md_exp, op_data, NULL, 0, &request);
1904 ptlrpc_req_finished(request);
1905 if (rc == -ENOENT) {
1907 /* Unlinked special device node? Or just a race?
1908 * Pretend we done everything. */
1909 if (!S_ISREG(inode->i_mode) &&
1910 !S_ISDIR(inode->i_mode)) {
1911 ia_valid = op_data->op_attr.ia_valid;
1912 op_data->op_attr.ia_valid &= ~TIMES_SET_FLAGS;
1913 rc = simple_setattr(&init_user_ns, dentry,
1915 op_data->op_attr.ia_valid = ia_valid;
1917 } else if (rc != -EPERM && rc != -EACCES && rc != -ETXTBSY) {
1918 CERROR("md_setattr fails: rc = %d\n", rc);
1923 rc = md_get_lustre_md(sbi->ll_md_exp, &request->rq_pill, sbi->ll_dt_exp,
1924 sbi->ll_md_exp, &md);
1926 ptlrpc_req_finished(request);
1930 ia_valid = op_data->op_attr.ia_valid;
1931 /* inode size will be in ll_setattr_ost, can't do it now since dirty
1932 * cache is not cleared yet. */
1933 op_data->op_attr.ia_valid &= ~(TIMES_SET_FLAGS | ATTR_SIZE);
1934 if (S_ISREG(inode->i_mode))
1936 rc = simple_setattr(&init_user_ns, dentry, &op_data->op_attr);
1937 if (S_ISREG(inode->i_mode))
1938 inode_unlock(inode);
1939 op_data->op_attr.ia_valid = ia_valid;
1941 rc = ll_update_inode(inode, &md);
1942 ptlrpc_req_finished(request);
1948 * Zero portion of page that is part of @inode.
1949 * This implies, if necessary:
1950 * - taking cl_lock on range corresponding to concerned page
1951 * - grabbing vm page
1952 * - associating cl_page
1953 * - proceeding to clio read
1954 * - zeroing range in page
1955 * - proceeding to cl_page flush
1956 * - releasing cl_lock
1958 * \param[in] inode inode
1959 * \param[in] index page index
1960 * \param[in] offset offset in page to start zero from
1961 * \param[in] len len to zero
1963 * \retval 0 on success
1964 * \retval negative errno on failure
1966 int ll_io_zero_page(struct inode *inode, pgoff_t index, pgoff_t offset,
1969 struct ll_inode_info *lli = ll_i2info(inode);
1970 struct cl_object *clob = lli->lli_clob;
1972 struct lu_env *env = NULL;
1973 struct cl_io *io = NULL;
1974 struct cl_page *clpage = NULL;
1975 struct page *vmpage = NULL;
1976 unsigned from = index << PAGE_SHIFT;
1977 struct cl_lock *lock = NULL;
1978 struct cl_lock_descr *descr = NULL;
1979 struct cl_2queue *queue = NULL;
1980 struct cl_sync_io *anchor = NULL;
1981 bool holdinglock = false;
1986 env = cl_env_get(&refcheck);
1988 RETURN(PTR_ERR(env));
1990 io = vvp_env_thread_io(env);
1992 rc = cl_io_rw_init(env, io, CIT_WRITE, from, PAGE_SIZE);
1996 lock = vvp_env_lock(env);
1997 descr = &lock->cll_descr;
1998 descr->cld_obj = io->ci_obj;
1999 descr->cld_start = cl_index(io->ci_obj, from);
2000 descr->cld_end = cl_index(io->ci_obj, from + PAGE_SIZE - 1);
2001 descr->cld_mode = CLM_WRITE;
2002 descr->cld_enq_flags = CEF_MUST | CEF_NONBLOCK;
2004 /* request lock for page */
2005 rc = cl_lock_request(env, io, lock);
2006 /* -ECANCELED indicates a matching lock with a different extent
2007 * was already present, and -EEXIST indicates a matching lock
2008 * on exactly the same extent was already present.
2009 * In both cases it means we are covered.
2011 if (rc == -ECANCELED || rc == -EEXIST)
2019 vmpage = grab_cache_page_nowait(inode->i_mapping, index);
2021 GOTO(rellock, rc = -EOPNOTSUPP);
2023 if (!PageDirty(vmpage)) {
2024 /* associate cl_page */
2025 clpage = cl_page_find(env, clob, vmpage->index,
2026 vmpage, CPT_CACHEABLE);
2028 GOTO(pagefini, rc = PTR_ERR(clpage));
2030 cl_page_assume(env, io, clpage);
2033 if (!PageUptodate(vmpage) && !PageDirty(vmpage) &&
2034 !PageWriteback(vmpage)) {
2036 /* Set PagePrivate2 to detect special case of empty page
2037 * in osc_brw_fini_request().
2038 * It is also used to tell ll_io_read_page() that we do not
2039 * want the vmpage to be unlocked.
2041 SetPagePrivate2(vmpage);
2042 rc = ll_io_read_page(env, io, clpage, NULL);
2043 if (!PagePrivate2(vmpage)) {
2044 /* PagePrivate2 was cleared in osc_brw_fini_request()
2045 * meaning we read an empty page. In this case, in order
2046 * to avoid allocating unnecessary block in truncated
2047 * file, we must not zero and write as below. Subsequent
2048 * server-side truncate will handle things correctly.
2050 cl_page_unassume(env, io, clpage);
2051 GOTO(clpfini, rc = 0);
2053 ClearPagePrivate2(vmpage);
2058 /* Thanks to PagePrivate2 flag, ll_io_read_page() did not unlock
2059 * the vmpage, so we are good to proceed and zero range in page.
2061 zero_user(vmpage, offset, len);
2063 if (holdinglock && clpage) {
2064 /* explicitly write newly modified page */
2065 queue = &io->ci_queue;
2066 cl_2queue_init(queue);
2067 anchor = &vvp_env_info(env)->vti_anchor;
2068 cl_sync_io_init(anchor, 1);
2069 clpage->cp_sync_io = anchor;
2070 cl_2queue_add(queue, clpage, true);
2071 rc = cl_io_submit_rw(env, io, CRT_WRITE, queue);
2073 GOTO(queuefini1, rc);
2074 rc = cl_sync_io_wait(env, anchor, 0);
2076 GOTO(queuefini2, rc);
2077 cl_page_assume(env, io, clpage);
2080 cl_2queue_discard(env, io, queue);
2082 cl_2queue_disown(env, io, queue);
2083 cl_2queue_fini(env, queue);
2088 cl_page_put(env, clpage);
2090 unlock_page(vmpage);
2094 cl_lock_release(env, lock);
2096 cl_io_fini(env, io);
2099 cl_env_put(env, &refcheck);
2105 * Get reference file from volatile file name.
2106 * Volatile file name may look like:
2107 * <parent>/LUSTRE_VOLATILE_HDR:<mdt_index>:<random>:fd=<fd>
2108 * where fd is opened descriptor of reference file.
2110 * \param[in] volatile_name volatile file name
2111 * \param[in] volatile_len volatile file name length
2112 * \param[out] ref_file pointer to struct file of reference file
2114 * \retval 0 on success
2115 * \retval negative errno on failure
2117 int volatile_ref_file(const char *volatile_name, int volatile_len,
2118 struct file **ref_file)
2120 char *p, *q, *fd_str;
2123 p = strnstr(volatile_name, ":fd=", volatile_len);
2124 if (!p || strlen(p + 4) == 0)
2127 q = strchrnul(p + 4, ':');
2128 fd_str = kstrndup(p + 4, q - p - 4, GFP_NOFS);
2131 rc = kstrtouint(fd_str, 10, &fd);
2136 *ref_file = fget(fd);
2142 /* If this inode has objects allocated to it (lsm != NULL), then the OST
2143 * object(s) determine the file size and mtime. Otherwise, the MDS will
2144 * keep these values until such a time that objects are allocated for it.
2145 * We do the MDS operations first, as it is checking permissions for us.
2146 * We don't to the MDS RPC if there is nothing that we want to store there,
2147 * otherwise there is no harm in updating mtime/atime on the MDS if we are
2148 * going to do an RPC anyways.
2150 * If we are doing a truncate, we will send the mtime and ctime updates
2151 * to the OST with the punch RPC, otherwise we do an explicit setattr RPC.
2152 * I don't believe it is possible to get e.g. ATTR_MTIME_SET and ATTR_SIZE
2155 * In case of HSMimport, we only set attr on MDS.
2157 int ll_setattr_raw(struct dentry *dentry, struct iattr *attr,
2158 enum op_xvalid xvalid, bool hsm_import)
2160 struct inode *inode = dentry->d_inode;
2161 struct ll_inode_info *lli = ll_i2info(inode);
2162 struct md_op_data *op_data = NULL;
2163 ktime_t kstart = ktime_get();
2168 CDEBUG(D_VFSTRACE, "%s: setattr inode "DFID"(%p) from %llu to %llu, "
2169 "valid %x, hsm_import %d\n",
2170 ll_i2sbi(inode)->ll_fsname, PFID(&lli->lli_fid),
2171 inode, i_size_read(inode), attr->ia_size, attr->ia_valid,
2174 if (attr->ia_valid & ATTR_SIZE) {
2175 /* Check new size against VFS/VM file size limit and rlimit */
2176 rc = inode_newsize_ok(inode, attr->ia_size);
2180 /* The maximum Lustre file size is variable, based on the
2181 * OST maximum object size and number of stripes. This
2182 * needs another check in addition to the VFS check above. */
2183 if (attr->ia_size > ll_file_maxbytes(inode)) {
2184 CDEBUG(D_INODE,"file "DFID" too large %llu > %llu\n",
2185 PFID(&lli->lli_fid), attr->ia_size,
2186 ll_file_maxbytes(inode));
2190 attr->ia_valid |= ATTR_MTIME | ATTR_CTIME;
2193 /* POSIX: check before ATTR_*TIME_SET set (from inode_change_ok) */
2194 if (attr->ia_valid & TIMES_SET_FLAGS) {
2195 if ((!uid_eq(current_fsuid(), inode->i_uid)) &&
2196 !capable(CAP_FOWNER))
2200 /* We mark all of the fields "set" so MDS/OST does not re-set them */
2201 if (!(xvalid & OP_XVALID_CTIME_SET) &&
2202 (attr->ia_valid & ATTR_CTIME)) {
2203 attr->ia_ctime = current_time(inode);
2204 xvalid |= OP_XVALID_CTIME_SET;
2206 if (!(attr->ia_valid & ATTR_ATIME_SET) &&
2207 (attr->ia_valid & ATTR_ATIME)) {
2208 attr->ia_atime = current_time(inode);
2209 attr->ia_valid |= ATTR_ATIME_SET;
2211 if (!(attr->ia_valid & ATTR_MTIME_SET) &&
2212 (attr->ia_valid & ATTR_MTIME)) {
2213 attr->ia_mtime = current_time(inode);
2214 attr->ia_valid |= ATTR_MTIME_SET;
2217 if (attr->ia_valid & (ATTR_MTIME | ATTR_CTIME))
2218 CDEBUG(D_INODE, "setting mtime %lld, ctime %lld, now = %lld\n",
2219 (s64)attr->ia_mtime.tv_sec, (s64)attr->ia_ctime.tv_sec,
2220 ktime_get_real_seconds());
2222 if (S_ISREG(inode->i_mode))
2223 inode_unlock(inode);
2225 /* We always do an MDS RPC, even if we're only changing the size;
2226 * only the MDS knows whether truncate() should fail with -ETXTBUSY */
2228 OBD_ALLOC_PTR(op_data);
2229 if (op_data == NULL)
2230 GOTO(out, rc = -ENOMEM);
2232 if (!hsm_import && attr->ia_valid & ATTR_SIZE) {
2233 /* If we are changing file size, file content is
2234 * modified, flag it.
2236 xvalid |= OP_XVALID_OWNEROVERRIDE;
2237 op_data->op_bias |= MDS_DATA_MODIFIED;
2238 clear_bit(LLIF_DATA_MODIFIED, &lli->lli_flags);
2241 if (attr->ia_valid & ATTR_FILE) {
2242 struct ll_file_data *fd = attr->ia_file->private_data;
2244 if (fd->fd_lease_och)
2245 op_data->op_bias |= MDS_TRUNC_KEEP_LEASE;
2248 op_data->op_attr = *attr;
2249 op_data->op_xvalid = xvalid;
2251 rc = ll_md_setattr(dentry, op_data);
2255 if (!S_ISREG(inode->i_mode) || hsm_import)
2258 if (attr->ia_valid & (ATTR_SIZE | ATTR_ATIME | ATTR_ATIME_SET |
2259 ATTR_MTIME | ATTR_MTIME_SET | ATTR_CTIME) ||
2260 xvalid & OP_XVALID_CTIME_SET) {
2261 bool cached = false;
2263 rc = pcc_inode_setattr(inode, attr, &cached);
2266 CERROR("%s: PCC inode "DFID" setattr failed: "
2268 ll_i2sbi(inode)->ll_fsname,
2269 PFID(&lli->lli_fid), rc);
2273 unsigned int flags = 0;
2275 /* For truncate and utimes sending attributes to OSTs,
2276 * setting mtime/atime to the past will be performed
2277 * under PW [0:EOF] extent lock (new_size:EOF for
2278 * truncate). It may seem excessive to send mtime/atime
2279 * updates to OSTs when not setting times to past, but
2280 * it is necessary due to possible time
2281 * de-synchronization between MDT inode and OST objects
2283 if (S_ISREG(inode->i_mode) && IS_ENCRYPTED(inode)) {
2284 xvalid |= OP_XVALID_FLAGS;
2285 flags = LUSTRE_ENCRYPT_FL;
2286 /* Call to ll_io_zero_page is not necessary if
2287 * truncating on PAGE_SIZE boundary, because
2288 * whole pages will be wiped.
2289 * In case of Direct IO, all we need is to set
2292 if (attr->ia_valid & ATTR_SIZE &&
2293 attr->ia_size & ~PAGE_MASK &&
2294 !(attr->ia_valid & ATTR_FILE &&
2295 attr->ia_file->f_flags & O_DIRECT)) {
2297 attr->ia_size & (PAGE_SIZE - 1);
2299 rc = ll_io_zero_page(inode,
2300 attr->ia_size >> PAGE_SHIFT,
2301 offset, PAGE_SIZE - offset);
2305 /* If encrypted volatile file without the key,
2306 * we need to fetch size from reference file,
2307 * and set it on OST objects. This happens when
2308 * migrating or extending an encrypted file
2311 if (filename_is_volatile(dentry->d_name.name,
2314 llcrypt_require_key(inode) == -ENOKEY) {
2315 struct file *ref_file;
2316 struct inode *ref_inode;
2317 struct ll_inode_info *ref_lli;
2318 struct cl_object *ref_obj;
2319 struct cl_attr ref_attr = { 0 };
2323 rc = volatile_ref_file(
2324 dentry->d_name.name,
2330 ref_inode = file_inode(ref_file);
2333 GOTO(out, rc = -EINVAL);
2336 env = cl_env_get(&refcheck);
2338 GOTO(out, rc = PTR_ERR(env));
2340 ref_lli = ll_i2info(ref_inode);
2341 ref_obj = ref_lli->lli_clob;
2342 cl_object_attr_lock(ref_obj);
2343 rc = cl_object_attr_get(env, ref_obj,
2345 cl_object_attr_unlock(ref_obj);
2346 cl_env_put(env, &refcheck);
2351 attr->ia_valid |= ATTR_SIZE;
2352 attr->ia_size = ref_attr.cat_size;
2355 rc = cl_setattr_ost(lli->lli_clob, attr, xvalid, flags);
2359 /* If the file was restored, it needs to set dirty flag.
2361 * We've already sent MDS_DATA_MODIFIED flag in
2362 * ll_md_setattr() for truncate. However, the MDT refuses to
2363 * set the HS_DIRTY flag on released files, so we have to set
2364 * it again if the file has been restored. Please check how
2365 * LLIF_DATA_MODIFIED is set in vvp_io_setattr_fini().
2367 * Please notice that if the file is not released, the previous
2368 * MDS_DATA_MODIFIED has taken effect and usually
2369 * LLIF_DATA_MODIFIED is not set(see vvp_io_setattr_fini()).
2370 * This way we can save an RPC for common open + trunc
2372 if (test_and_clear_bit(LLIF_DATA_MODIFIED, &lli->lli_flags)) {
2373 struct hsm_state_set hss = {
2374 .hss_valid = HSS_SETMASK,
2375 .hss_setmask = HS_DIRTY,
2379 rc2 = ll_hsm_state_set(inode, &hss);
2380 /* truncate and write can happen at the same time, so that
2381 * the file can be set modified even though the file is not
2382 * restored from released state, and ll_hsm_state_set() is
2383 * not applicable for the file, and rc2 < 0 is normal in this
2386 CDEBUG(D_INFO, DFID "HSM set dirty failed: rc2 = %d\n",
2387 PFID(ll_inode2fid(inode)), rc2);
2392 if (op_data != NULL)
2393 ll_finish_md_op_data(op_data);
2395 if (S_ISREG(inode->i_mode)) {
2397 if ((attr->ia_valid & ATTR_SIZE) && !hsm_import)
2398 inode_dio_wait(inode);
2399 /* Once we've got the i_mutex, it's safe to set the S_NOSEC
2400 * flag. ll_update_inode (called from ll_md_setattr), clears
2401 * inode flags, so there is a gap where S_NOSEC is not set.
2402 * This can cause a writer to take the i_mutex unnecessarily,
2403 * but this is safe to do and should be rare. */
2404 inode_has_no_xattr(inode);
2408 ll_stats_ops_tally(ll_i2sbi(inode), attr->ia_valid & ATTR_SIZE ?
2409 LPROC_LL_TRUNC : LPROC_LL_SETATTR,
2410 ktime_us_delta(ktime_get(), kstart));
2415 int ll_setattr(struct user_namespace *mnt_userns, struct dentry *de,
2418 int mode = de->d_inode->i_mode;
2419 enum op_xvalid xvalid = 0;
2422 rc = llcrypt_prepare_setattr(de, attr);
2426 if ((attr->ia_valid & (ATTR_CTIME|ATTR_SIZE|ATTR_MODE)) ==
2427 (ATTR_CTIME|ATTR_SIZE|ATTR_MODE))
2428 xvalid |= OP_XVALID_OWNEROVERRIDE;
2430 if (((attr->ia_valid & (ATTR_MODE|ATTR_FORCE|ATTR_SIZE)) ==
2431 (ATTR_SIZE|ATTR_MODE)) &&
2432 (((mode & S_ISUID) && !(attr->ia_mode & S_ISUID)) ||
2433 (((mode & (S_ISGID|S_IXGRP)) == (S_ISGID|S_IXGRP)) &&
2434 !(attr->ia_mode & S_ISGID))))
2435 attr->ia_valid |= ATTR_FORCE;
2437 if ((attr->ia_valid & ATTR_MODE) &&
2439 !(attr->ia_mode & S_ISUID) &&
2440 !(attr->ia_valid & ATTR_KILL_SUID))
2441 attr->ia_valid |= ATTR_KILL_SUID;
2443 if ((attr->ia_valid & ATTR_MODE) &&
2444 ((mode & (S_ISGID|S_IXGRP)) == (S_ISGID|S_IXGRP)) &&
2445 !(attr->ia_mode & S_ISGID) &&
2446 !(attr->ia_valid & ATTR_KILL_SGID))
2447 attr->ia_valid |= ATTR_KILL_SGID;
2449 return ll_setattr_raw(de, attr, xvalid, false);
2452 int ll_statfs_internal(struct ll_sb_info *sbi, struct obd_statfs *osfs,
2455 struct obd_statfs obd_osfs = { 0 };
2460 max_age = ktime_get_seconds() - sbi->ll_statfs_max_age;
2462 if (test_bit(LL_SBI_LAZYSTATFS, sbi->ll_flags))
2463 flags |= OBD_STATFS_NODELAY;
2465 rc = obd_statfs(NULL, sbi->ll_md_exp, osfs, max_age, flags);
2469 osfs->os_type = LL_SUPER_MAGIC;
2471 CDEBUG(D_SUPER, "MDC blocks %llu/%llu objects %llu/%llu\n",
2472 osfs->os_bavail, osfs->os_blocks, osfs->os_ffree, osfs->os_files);
2474 if (osfs->os_state & OS_STATFS_SUM)
2477 rc = obd_statfs(NULL, sbi->ll_dt_exp, &obd_osfs, max_age, flags);
2478 if (rc) /* Possibly a filesystem with no OSTs. Report MDT totals. */
2481 CDEBUG(D_SUPER, "OSC blocks %llu/%llu objects %llu/%llu\n",
2482 obd_osfs.os_bavail, obd_osfs.os_blocks, obd_osfs.os_ffree,
2485 osfs->os_bsize = obd_osfs.os_bsize;
2486 osfs->os_blocks = obd_osfs.os_blocks;
2487 osfs->os_bfree = obd_osfs.os_bfree;
2488 osfs->os_bavail = obd_osfs.os_bavail;
2490 /* If we have _some_ OSTs, but don't have as many free objects on the
2491 * OSTs as inodes on the MDTs, reduce the reported number of inodes
2492 * to compensate, so that the "inodes in use" number is correct.
2493 * This should be kept in sync with lod_statfs() behaviour.
2495 if (obd_osfs.os_files && obd_osfs.os_ffree < osfs->os_ffree) {
2496 osfs->os_files = (osfs->os_files - osfs->os_ffree) +
2498 osfs->os_ffree = obd_osfs.os_ffree;
2505 static int ll_statfs_project(struct inode *inode, struct kstatfs *sfs)
2507 struct if_quotactl qctl = {
2508 .qc_cmd = LUSTRE_Q_GETQUOTA,
2509 .qc_type = PRJQUOTA,
2510 .qc_valid = QC_GENERAL,
2512 u64 limit, curblock;
2515 qctl.qc_id = ll_i2info(inode)->lli_projid;
2516 ret = quotactl_ioctl(inode->i_sb, &qctl);
2518 /* ignore errors if project ID does not have
2519 * a quota limit or feature unsupported.
2521 if (ret == -ESRCH || ret == -EOPNOTSUPP)
2526 limit = ((qctl.qc_dqblk.dqb_bsoftlimit ?
2527 qctl.qc_dqblk.dqb_bsoftlimit :
2528 qctl.qc_dqblk.dqb_bhardlimit) * 1024) / sfs->f_bsize;
2529 if (limit && sfs->f_blocks > limit) {
2530 curblock = (qctl.qc_dqblk.dqb_curspace +
2531 sfs->f_bsize - 1) / sfs->f_bsize;
2532 sfs->f_blocks = limit;
2533 sfs->f_bfree = sfs->f_bavail =
2534 (sfs->f_blocks > curblock) ?
2535 (sfs->f_blocks - curblock) : 0;
2538 limit = qctl.qc_dqblk.dqb_isoftlimit ?
2539 qctl.qc_dqblk.dqb_isoftlimit :
2540 qctl.qc_dqblk.dqb_ihardlimit;
2541 if (limit && sfs->f_files > limit) {
2542 sfs->f_files = limit;
2543 sfs->f_ffree = (sfs->f_files >
2544 qctl.qc_dqblk.dqb_curinodes) ?
2545 (sfs->f_files - qctl.qc_dqblk.dqb_curinodes) : 0;
2551 int ll_statfs(struct dentry *de, struct kstatfs *sfs)
2553 struct super_block *sb = de->d_sb;
2554 struct obd_statfs osfs;
2555 __u64 fsid = huge_encode_dev(sb->s_dev);
2556 ktime_t kstart = ktime_get();
2559 CDEBUG(D_VFSTRACE, "VFS Op:sb=%s (%p)\n", sb->s_id, sb);
2561 /* Some amount of caching on the client is allowed */
2562 rc = ll_statfs_internal(ll_s2sbi(sb), &osfs, OBD_STATFS_SUM);
2566 statfs_unpack(sfs, &osfs);
2568 /* We need to downshift for all 32-bit kernels, because we can't
2569 * tell if the kernel is being called via sys_statfs64() or not.
2570 * Stop before overflowing f_bsize - in which case it is better
2571 * to just risk EOVERFLOW if caller is using old sys_statfs(). */
2572 if (sizeof(long) < 8) {
2573 while (osfs.os_blocks > ~0UL && sfs->f_bsize < 0x40000000) {
2576 osfs.os_blocks >>= 1;
2577 osfs.os_bfree >>= 1;
2578 osfs.os_bavail >>= 1;
2582 sfs->f_blocks = osfs.os_blocks;
2583 sfs->f_bfree = osfs.os_bfree;
2584 sfs->f_bavail = osfs.os_bavail;
2585 sfs->f_fsid.val[0] = (__u32)fsid;
2586 sfs->f_fsid.val[1] = (__u32)(fsid >> 32);
2587 if (ll_i2info(de->d_inode)->lli_projid)
2588 return ll_statfs_project(de->d_inode, sfs);
2590 ll_stats_ops_tally(ll_s2sbi(sb), LPROC_LL_STATFS,
2591 ktime_us_delta(ktime_get(), kstart));
2596 void ll_inode_size_lock(struct inode *inode)
2598 struct ll_inode_info *lli;
2600 LASSERT(!S_ISDIR(inode->i_mode));
2602 lli = ll_i2info(inode);
2603 mutex_lock(&lli->lli_size_mutex);
2606 void ll_inode_size_unlock(struct inode *inode)
2608 struct ll_inode_info *lli;
2610 lli = ll_i2info(inode);
2611 mutex_unlock(&lli->lli_size_mutex);
2614 void ll_update_inode_flags(struct inode *inode, unsigned int ext_flags)
2616 /* do not clear encryption flag */
2617 ext_flags |= ll_inode_to_ext_flags(inode->i_flags) & LUSTRE_ENCRYPT_FL;
2618 inode->i_flags = ll_ext_to_inode_flags(ext_flags);
2619 if (ext_flags & LUSTRE_PROJINHERIT_FL)
2620 set_bit(LLIF_PROJECT_INHERIT, &ll_i2info(inode)->lli_flags);
2622 clear_bit(LLIF_PROJECT_INHERIT, &ll_i2info(inode)->lli_flags);
2625 int ll_update_inode(struct inode *inode, struct lustre_md *md)
2627 struct ll_inode_info *lli = ll_i2info(inode);
2628 struct mdt_body *body = md->body;
2629 struct ll_sb_info *sbi = ll_i2sbi(inode);
2633 if (body->mbo_valid & OBD_MD_FLEASIZE) {
2634 rc = cl_file_inode_init(inode, md);
2639 if (S_ISDIR(inode->i_mode)) {
2640 rc = ll_update_lsm_md(inode, md);
2645 if (body->mbo_valid & OBD_MD_FLACL)
2646 lli_replace_acl(lli, md);
2648 api32 = test_bit(LL_SBI_32BIT_API, sbi->ll_flags);
2649 inode->i_ino = cl_fid_build_ino(&body->mbo_fid1, api32);
2650 inode->i_generation = cl_fid_build_gen(&body->mbo_fid1);
2652 if (body->mbo_valid & OBD_MD_FLATIME) {
2653 if (body->mbo_atime > inode->i_atime.tv_sec)
2654 inode->i_atime.tv_sec = body->mbo_atime;
2655 lli->lli_atime = body->mbo_atime;
2658 if (body->mbo_valid & OBD_MD_FLMTIME) {
2659 if (body->mbo_mtime > inode->i_mtime.tv_sec) {
2661 "setting ino %lu mtime from %lld to %llu\n",
2662 inode->i_ino, (s64)inode->i_mtime.tv_sec,
2664 inode->i_mtime.tv_sec = body->mbo_mtime;
2666 lli->lli_mtime = body->mbo_mtime;
2669 if (body->mbo_valid & OBD_MD_FLCTIME) {
2670 if (body->mbo_ctime > inode->i_ctime.tv_sec)
2671 inode->i_ctime.tv_sec = body->mbo_ctime;
2672 lli->lli_ctime = body->mbo_ctime;
2675 if (body->mbo_valid & OBD_MD_FLBTIME)
2676 lli->lli_btime = body->mbo_btime;
2678 /* Clear i_flags to remove S_NOSEC before permissions are updated */
2679 if (body->mbo_valid & OBD_MD_FLFLAGS)
2680 ll_update_inode_flags(inode, body->mbo_flags);
2681 if (body->mbo_valid & OBD_MD_FLMODE)
2682 inode->i_mode = (inode->i_mode & S_IFMT) |
2683 (body->mbo_mode & ~S_IFMT);
2685 if (body->mbo_valid & OBD_MD_FLTYPE)
2686 inode->i_mode = (inode->i_mode & ~S_IFMT) |
2687 (body->mbo_mode & S_IFMT);
2689 LASSERT(inode->i_mode != 0);
2690 if (body->mbo_valid & OBD_MD_FLUID)
2691 inode->i_uid = make_kuid(&init_user_ns, body->mbo_uid);
2692 if (body->mbo_valid & OBD_MD_FLGID)
2693 inode->i_gid = make_kgid(&init_user_ns, body->mbo_gid);
2694 if (body->mbo_valid & OBD_MD_FLPROJID)
2695 lli->lli_projid = body->mbo_projid;
2696 if (body->mbo_valid & OBD_MD_FLNLINK) {
2697 spin_lock(&inode->i_lock);
2698 set_nlink(inode, body->mbo_nlink);
2699 spin_unlock(&inode->i_lock);
2701 if (body->mbo_valid & OBD_MD_FLRDEV)
2702 inode->i_rdev = old_decode_dev(body->mbo_rdev);
2704 if (body->mbo_valid & OBD_MD_FLID) {
2705 /* FID shouldn't be changed! */
2706 if (fid_is_sane(&lli->lli_fid)) {
2707 LASSERTF(lu_fid_eq(&lli->lli_fid, &body->mbo_fid1),
2708 "Trying to change FID "DFID
2709 " to the "DFID", inode "DFID"(%p)\n",
2710 PFID(&lli->lli_fid), PFID(&body->mbo_fid1),
2711 PFID(ll_inode2fid(inode)), inode);
2713 lli->lli_fid = body->mbo_fid1;
2717 LASSERT(fid_seq(&lli->lli_fid) != 0);
2719 /* In case of encrypted file without the key, please do not lose
2720 * clear text size stored into lli_lazysize in ll_merge_attr(),
2721 * we will need it in ll_prepare_close().
2723 if (lli->lli_attr_valid & OBD_MD_FLLAZYSIZE && lli->lli_lazysize &&
2724 llcrypt_require_key(inode) == -ENOKEY)
2725 lli->lli_attr_valid = body->mbo_valid | OBD_MD_FLLAZYSIZE;
2727 lli->lli_attr_valid = body->mbo_valid;
2728 if (body->mbo_valid & OBD_MD_FLSIZE) {
2729 i_size_write(inode, body->mbo_size);
2731 CDEBUG(D_VFSTRACE, "inode="DFID", updating i_size %llu\n",
2732 PFID(ll_inode2fid(inode)),
2733 (unsigned long long)body->mbo_size);
2735 if (body->mbo_valid & OBD_MD_FLBLOCKS)
2736 inode->i_blocks = body->mbo_blocks;
2738 if (body->mbo_valid & OBD_MD_FLLAZYSIZE)
2739 lli->lli_lazysize = body->mbo_size;
2740 if (body->mbo_valid & OBD_MD_FLLAZYBLOCKS)
2741 lli->lli_lazyblocks = body->mbo_blocks;
2744 if (body->mbo_valid & OBD_MD_TSTATE) {
2745 /* Set LLIF_FILE_RESTORING if restore ongoing and
2746 * clear it when done to ensure to start again
2747 * glimpsing updated attrs
2749 if (body->mbo_t_state & MS_RESTORE)
2750 set_bit(LLIF_FILE_RESTORING, &lli->lli_flags);
2752 clear_bit(LLIF_FILE_RESTORING, &lli->lli_flags);
2758 /* update directory depth to ROOT, called after LOOKUP lock is fetched. */
2759 void ll_update_dir_depth(struct inode *dir, struct inode *inode)
2761 struct ll_inode_info *lli;
2763 if (!S_ISDIR(inode->i_mode))
2769 lli = ll_i2info(inode);
2770 lli->lli_dir_depth = ll_i2info(dir)->lli_dir_depth + 1;
2771 CDEBUG(D_INODE, DFID" depth %hu\n",
2772 PFID(&lli->lli_fid), lli->lli_dir_depth);
2775 void ll_truncate_inode_pages_final(struct inode *inode)
2777 struct address_space *mapping = &inode->i_data;
2778 unsigned long nrpages;
2779 unsigned long flags;
2781 truncate_inode_pages_final(mapping);
2783 /* Workaround for LU-118: Note nrpages may not be totally updated when
2784 * truncate_inode_pages() returns, as there can be a page in the process
2785 * of deletion (inside __delete_from_page_cache()) in the specified
2786 * range. Thus mapping->nrpages can be non-zero when this function
2787 * returns even after truncation of the whole mapping. Only do this if
2788 * npages isn't already zero.
2790 nrpages = mapping->nrpages;
2792 ll_xa_lock_irqsave(&mapping->i_pages, flags);
2793 nrpages = mapping->nrpages;
2794 ll_xa_unlock_irqrestore(&mapping->i_pages, flags);
2795 } /* Workaround end */
2797 LASSERTF(nrpages == 0, "%s: inode="DFID"(%p) nrpages=%lu, "
2798 "see https://jira.whamcloud.com/browse/LU-118\n",
2799 ll_i2sbi(inode)->ll_fsname,
2800 PFID(ll_inode2fid(inode)), inode, nrpages);
2803 int ll_read_inode2(struct inode *inode, void *opaque)
2805 struct lustre_md *md = opaque;
2806 struct ll_inode_info *lli = ll_i2info(inode);
2810 CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p)\n",
2811 PFID(&lli->lli_fid), inode);
2813 /* Core attributes from the MDS first. This is a new inode, and
2814 * the VFS doesn't zero times in the core inode so we have to do
2815 * it ourselves. They will be overwritten by either MDS or OST
2816 * attributes - we just need to make sure they aren't newer.
2818 inode->i_mtime.tv_sec = 0;
2819 inode->i_atime.tv_sec = 0;
2820 inode->i_ctime.tv_sec = 0;
2822 rc = ll_update_inode(inode, md);
2826 /* OIDEBUG(inode); */
2828 #ifdef HAVE_BACKING_DEV_INFO
2829 /* initializing backing dev info. */
2830 inode->i_mapping->backing_dev_info = &s2lsi(inode->i_sb)->lsi_bdi;
2832 if (S_ISREG(inode->i_mode)) {
2833 struct ll_sb_info *sbi = ll_i2sbi(inode);
2834 inode->i_op = &ll_file_inode_operations;
2835 inode->i_fop = sbi->ll_fop;
2836 inode->i_mapping->a_ops = (struct address_space_operations *)&ll_aops;
2838 } else if (S_ISDIR(inode->i_mode)) {
2839 inode->i_op = &ll_dir_inode_operations;
2840 inode->i_fop = &ll_dir_operations;
2842 } else if (S_ISLNK(inode->i_mode)) {
2843 inode->i_op = &ll_fast_symlink_inode_operations;
2846 inode->i_op = &ll_special_inode_operations;
2848 init_special_inode(inode, inode->i_mode,
2857 void ll_delete_inode(struct inode *inode)
2859 struct ll_inode_info *lli = ll_i2info(inode);
2862 if (S_ISREG(inode->i_mode) && lli->lli_clob != NULL) {
2863 /* It is last chance to write out dirty pages,
2864 * otherwise we may lose data while umount.
2866 * If i_nlink is 0 then just discard data. This is safe because
2867 * local inode gets i_nlink 0 from server only for the last
2868 * unlink, so that file is not opened somewhere else
2870 cl_sync_file_range(inode, 0, OBD_OBJECT_EOF, inode->i_nlink ?
2871 CL_FSYNC_LOCAL : CL_FSYNC_DISCARD, 1);
2874 ll_truncate_inode_pages_final(inode);
2875 ll_clear_inode(inode);
2881 int ll_iocontrol(struct inode *inode, struct file *file,
2882 unsigned int cmd, unsigned long arg)
2884 struct ll_sb_info *sbi = ll_i2sbi(inode);
2885 struct ptlrpc_request *req = NULL;
2890 case FS_IOC_GETFLAGS: {
2891 struct mdt_body *body;
2892 struct md_op_data *op_data;
2894 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL,
2895 0, 0, LUSTRE_OPC_ANY,
2897 if (IS_ERR(op_data))
2898 RETURN(PTR_ERR(op_data));
2900 op_data->op_valid = OBD_MD_FLFLAGS;
2901 rc = md_getattr(sbi->ll_md_exp, op_data, &req);
2902 ll_finish_md_op_data(op_data);
2904 CERROR("%s: failure inode "DFID": rc = %d\n",
2905 sbi->ll_md_exp->exp_obd->obd_name,
2906 PFID(ll_inode2fid(inode)), rc);
2910 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
2912 flags = body->mbo_flags;
2914 ptlrpc_req_finished(req);
2916 RETURN(put_user(flags, (int __user *)arg));
2918 case FS_IOC_SETFLAGS: {
2920 struct md_op_data *op_data;
2921 struct cl_object *obj;
2922 struct fsxattr fa = { 0 };
2924 if (get_user(flags, (int __user *)arg))
2927 fa.fsx_projid = ll_i2info(inode)->lli_projid;
2928 if (flags & LUSTRE_PROJINHERIT_FL)
2929 fa.fsx_xflags = FS_XFLAG_PROJINHERIT;
2931 rc = ll_ioctl_check_project(inode, fa.fsx_xflags,
2936 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
2937 LUSTRE_OPC_ANY, NULL);
2938 if (IS_ERR(op_data))
2939 RETURN(PTR_ERR(op_data));
2941 op_data->op_attr_flags = flags;
2942 op_data->op_xvalid |= OP_XVALID_FLAGS;
2943 rc = md_setattr(sbi->ll_md_exp, op_data, NULL, 0, &req);
2944 ll_finish_md_op_data(op_data);
2945 ptlrpc_req_finished(req);
2949 ll_update_inode_flags(inode, flags);
2951 obj = ll_i2info(inode)->lli_clob;
2955 OBD_ALLOC_PTR(attr);
2959 rc = cl_setattr_ost(obj, attr, OP_XVALID_FLAGS, flags);
2971 int ll_flush_ctx(struct inode *inode)
2973 struct ll_sb_info *sbi = ll_i2sbi(inode);
2975 CDEBUG(D_SEC, "flush context for user %d\n",
2976 from_kuid(&init_user_ns, current_uid()));
2978 obd_set_info_async(NULL, sbi->ll_md_exp,
2979 sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
2981 obd_set_info_async(NULL, sbi->ll_dt_exp,
2982 sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
2987 /* umount -f client means force down, don't save state */
2988 void ll_umount_begin(struct super_block *sb)
2990 struct ll_sb_info *sbi = ll_s2sbi(sb);
2991 struct obd_device *obd;
2992 struct obd_ioctl_data *ioc_data;
2996 CDEBUG(D_VFSTRACE, "VFS Op: superblock %p count %d active %d\n", sb,
2997 sb->s_count, atomic_read(&sb->s_active));
2999 obd = class_exp2obd(sbi->ll_md_exp);
3001 CERROR("Invalid MDC connection handle %#llx\n",
3002 sbi->ll_md_exp->exp_handle.h_cookie);
3008 obd = class_exp2obd(sbi->ll_dt_exp);
3010 CERROR("Invalid LOV connection handle %#llx\n",
3011 sbi->ll_dt_exp->exp_handle.h_cookie);
3017 OBD_ALLOC_PTR(ioc_data);
3019 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_md_exp,
3020 sizeof *ioc_data, ioc_data, NULL);
3022 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_dt_exp,
3023 sizeof *ioc_data, ioc_data, NULL);
3025 OBD_FREE_PTR(ioc_data);
3028 /* Really, we'd like to wait until there are no requests outstanding,
3029 * and then continue. For now, we just periodically checking for vfs
3030 * to decrement mnt_cnt and hope to finish it within 10sec.
3034 !may_umount(sbi->ll_mnt.mnt)) {
3042 int ll_remount_fs(struct super_block *sb, int *flags, char *data)
3044 struct ll_sb_info *sbi = ll_s2sbi(sb);
3045 char *profilenm = get_profile_name(sb);
3049 if ((*flags & MS_RDONLY) != (sb->s_flags & SB_RDONLY)) {
3050 read_only = *flags & MS_RDONLY;
3051 err = obd_set_info_async(NULL, sbi->ll_md_exp,
3052 sizeof(KEY_READ_ONLY),
3053 KEY_READ_ONLY, sizeof(read_only),
3056 LCONSOLE_WARN("Failed to remount %s %s (%d)\n",
3057 profilenm, read_only ?
3058 "read-only" : "read-write", err);
3063 sb->s_flags |= SB_RDONLY;
3065 sb->s_flags &= ~SB_RDONLY;
3067 if (test_bit(LL_SBI_VERBOSE, sbi->ll_flags))
3068 LCONSOLE_WARN("Remounted %s %s\n", profilenm,
3069 read_only ? "read-only" : "read-write");
3075 * Cleanup the open handle that is cached on MDT-side.
3077 * For open case, the client side open handling thread may hit error
3078 * after the MDT grant the open. Under such case, the client should
3079 * send close RPC to the MDT as cleanup; otherwise, the open handle
3080 * on the MDT will be leaked there until the client umount or evicted.
3082 * In further, if someone unlinked the file, because the open handle
3083 * holds the reference on such file/object, then it will block the
3084 * subsequent threads that want to locate such object via FID.
3086 * \param[in] sb super block for this file-system
3087 * \param[in] open_req pointer to the original open request
3089 void ll_open_cleanup(struct super_block *sb, struct req_capsule *pill)
3091 struct mdt_body *body;
3092 struct md_op_data *op_data;
3093 struct ptlrpc_request *close_req = NULL;
3094 struct obd_export *exp = ll_s2sbi(sb)->ll_md_exp;
3097 body = req_capsule_server_get(pill, &RMF_MDT_BODY);
3098 OBD_ALLOC_PTR(op_data);
3099 if (op_data == NULL) {
3100 CWARN("%s: cannot allocate op_data to release open handle for "
3101 DFID"\n", ll_s2sbi(sb)->ll_fsname, PFID(&body->mbo_fid1));
3106 op_data->op_fid1 = body->mbo_fid1;
3107 op_data->op_open_handle = body->mbo_open_handle;
3108 op_data->op_mod_time = ktime_get_real_seconds();
3109 md_close(exp, op_data, NULL, &close_req);
3110 ptlrpc_req_finished(close_req);
3111 ll_finish_md_op_data(op_data);
3116 int ll_prep_inode(struct inode **inode, struct req_capsule *pill,
3117 struct super_block *sb, struct lookup_intent *it)
3119 struct ll_sb_info *sbi = NULL;
3120 struct lustre_md md = { NULL };
3121 bool default_lmv_deleted = false;
3126 LASSERT(*inode || sb);
3127 sbi = sb ? ll_s2sbi(sb) : ll_i2sbi(*inode);
3128 rc = md_get_lustre_md(sbi->ll_md_exp, pill, sbi->ll_dt_exp,
3129 sbi->ll_md_exp, &md);
3134 * clear default_lmv only if intent_getattr reply doesn't contain it.
3135 * but it needs to be done after iget, check this early because
3136 * ll_update_lsm_md() may change md.
3138 if (it && (it->it_op & (IT_LOOKUP | IT_GETATTR)) &&
3139 S_ISDIR(md.body->mbo_mode) && !md.default_lmv)
3140 default_lmv_deleted = true;
3143 rc = ll_update_inode(*inode, &md);
3147 bool api32 = test_bit(LL_SBI_32BIT_API, sbi->ll_flags);
3148 struct lu_fid *fid1 = &md.body->mbo_fid1;
3150 LASSERT(sb != NULL);
3153 * At this point server returns to client's same fid as client
3154 * generated for creating. So using ->fid1 is okay here.
3156 if (!fid_is_sane(fid1)) {
3157 CERROR("%s: Fid is insane "DFID"\n",
3158 sbi->ll_fsname, PFID(fid1));
3159 GOTO(out, rc = -EINVAL);
3162 *inode = ll_iget(sb, cl_fid_build_ino(fid1, api32), &md);
3163 if (IS_ERR(*inode)) {
3165 rc = IS_ERR(*inode) ? PTR_ERR(*inode) : -ENOMEM;
3167 CERROR("new_inode -fatal: rc %d\n", rc);
3172 /* Handling piggyback layout lock.
3173 * Layout lock can be piggybacked by getattr and open request.
3174 * The lsm can be applied to inode only if it comes with a layout lock
3175 * otherwise correct layout may be overwritten, for example:
3176 * 1. proc1: mdt returns a lsm but not granting layout
3177 * 2. layout was changed by another client
3178 * 3. proc2: refresh layout and layout lock granted
3179 * 4. proc1: to apply a stale layout */
3180 if (it != NULL && it->it_lock_mode != 0) {
3181 struct lustre_handle lockh;
3182 struct ldlm_lock *lock;
3184 lockh.cookie = it->it_lock_handle;
3185 lock = ldlm_handle2lock(&lockh);
3186 LASSERT(lock != NULL);
3187 if (ldlm_has_layout(lock)) {
3188 struct cl_object_conf conf;
3190 memset(&conf, 0, sizeof(conf));
3191 conf.coc_opc = OBJECT_CONF_SET;
3192 conf.coc_inode = *inode;
3193 conf.coc_lock = lock;
3194 conf.u.coc_layout = md.layout;
3195 (void)ll_layout_conf(*inode, &conf);
3197 LDLM_LOCK_PUT(lock);
3200 if (default_lmv_deleted)
3201 ll_update_default_lsm_md(*inode, &md);
3203 /* we may want to apply some policy for foreign file/dir */
3204 if (ll_sbi_has_foreign_symlink(sbi)) {
3205 rc = ll_manage_foreign(*inode, &md);
3213 /* cleanup will be done if necessary */
3214 md_free_lustre_md(sbi->ll_md_exp, &md);
3216 if (rc != 0 && it != NULL && it->it_op & IT_OPEN) {
3217 ll_intent_drop_lock(it);
3218 ll_open_cleanup(sb != NULL ? sb : (*inode)->i_sb, pill);
3224 int ll_obd_statfs(struct inode *inode, void __user *arg)
3226 struct ll_sb_info *sbi = NULL;
3227 struct obd_export *exp;
3228 struct obd_ioctl_data *data = NULL;
3233 sbi = ll_i2sbi(inode);
3235 GOTO(out_statfs, rc = -EINVAL);
3237 rc = obd_ioctl_getdata(&data, &len, arg);
3239 GOTO(out_statfs, rc);
3241 if (!data->ioc_inlbuf1 || !data->ioc_inlbuf2 ||
3242 !data->ioc_pbuf1 || !data->ioc_pbuf2)
3243 GOTO(out_statfs, rc = -EINVAL);
3245 if (data->ioc_inllen1 != sizeof(__u32) ||
3246 data->ioc_inllen2 != sizeof(__u32) ||
3247 data->ioc_plen1 != sizeof(struct obd_statfs) ||
3248 data->ioc_plen2 != sizeof(struct obd_uuid))
3249 GOTO(out_statfs, rc = -EINVAL);
3251 memcpy(&type, data->ioc_inlbuf1, sizeof(__u32));
3252 if (type & LL_STATFS_LMV)
3253 exp = sbi->ll_md_exp;
3254 else if (type & LL_STATFS_LOV)
3255 exp = sbi->ll_dt_exp;
3257 GOTO(out_statfs, rc = -ENODEV);
3259 rc = obd_iocontrol(IOC_OBD_STATFS, exp, len, data, NULL);
3261 GOTO(out_statfs, rc);
3263 OBD_FREE_LARGE(data, len);
3268 * this is normally called in ll_fini_md_op_data(), but sometimes it needs to
3269 * be called early to avoid deadlock.
3271 void ll_unlock_md_op_lsm(struct md_op_data *op_data)
3273 if (op_data->op_mea2_sem) {
3274 up_read_non_owner(op_data->op_mea2_sem);
3275 op_data->op_mea2_sem = NULL;
3278 if (op_data->op_mea1_sem) {
3279 up_read_non_owner(op_data->op_mea1_sem);
3280 op_data->op_mea1_sem = NULL;
3284 /* this function prepares md_op_data hint for passing it down to MD stack. */
3285 struct md_op_data *ll_prep_md_op_data(struct md_op_data *op_data,
3286 struct inode *i1, struct inode *i2,
3287 const char *name, size_t namelen,
3288 __u32 mode, enum md_op_code opc,
3291 struct llcrypt_name fname = { 0 };
3294 LASSERT(i1 != NULL);
3297 /* Do not reuse namelen for something else. */
3299 return ERR_PTR(-EINVAL);
3301 if ((!IS_ENCRYPTED(i1) ||
3302 (opc != LUSTRE_OPC_LOOKUP && opc != LUSTRE_OPC_CREATE)) &&
3303 namelen > ll_i2sbi(i1)->ll_namelen)
3304 return ERR_PTR(-ENAMETOOLONG);
3306 /* "/" is not valid name, but it's allowed */
3307 if (!lu_name_is_valid_2(name, namelen) &&
3308 strncmp("/", name, namelen) != 0)
3309 return ERR_PTR(-EINVAL);
3312 if (op_data == NULL)
3313 OBD_ALLOC_PTR(op_data);
3315 if (op_data == NULL)
3316 return ERR_PTR(-ENOMEM);
3318 ll_i2gids(op_data->op_suppgids, i1, i2);
3319 /* If the client is using a subdir mount and looks at what it sees as
3320 * /.fscrypt, interpret it as the .fscrypt dir at the root of the fs.
3322 if (unlikely(i1->i_sb && i1->i_sb->s_root && is_root_inode(i1) &&
3323 !fid_is_root(ll_inode2fid(i1)) &&
3324 name && namelen == strlen(dot_fscrypt_name) &&
3325 strncmp(name, dot_fscrypt_name, namelen) == 0))
3326 lu_root_fid(&op_data->op_fid1);
3328 op_data->op_fid1 = *ll_inode2fid(i1);
3330 if (S_ISDIR(i1->i_mode)) {
3331 down_read_non_owner(&ll_i2info(i1)->lli_lsm_sem);
3332 op_data->op_mea1_sem = &ll_i2info(i1)->lli_lsm_sem;
3333 op_data->op_mea1 = ll_i2info(i1)->lli_lsm_md;
3334 op_data->op_default_mea1 = ll_i2info(i1)->lli_default_lsm_md;
3338 op_data->op_fid2 = *ll_inode2fid(i2);
3339 if (S_ISDIR(i2->i_mode)) {
3341 /* i2 is typically a child of i1, and MUST be
3342 * further from the root to avoid deadlocks.
3344 down_read_non_owner(&ll_i2info(i2)->lli_lsm_sem);
3345 op_data->op_mea2_sem =
3346 &ll_i2info(i2)->lli_lsm_sem;
3348 op_data->op_mea2 = ll_i2info(i2)->lli_lsm_md;
3351 fid_zero(&op_data->op_fid2);
3354 if (test_bit(LL_SBI_64BIT_HASH, ll_i2sbi(i1)->ll_flags))
3355 op_data->op_cli_flags |= CLI_HASH64;
3357 if (ll_need_32bit_api(ll_i2sbi(i1)))
3358 op_data->op_cli_flags |= CLI_API32;
3360 if ((i2 && is_root_inode(i2)) ||
3361 opc == LUSTRE_OPC_LOOKUP || opc == LUSTRE_OPC_CREATE) {
3362 /* In case of lookup, ll_setup_filename() has already been
3363 * called in ll_lookup_it(), so just take provided name.
3364 * Also take provided name if we are dealing with root inode.
3366 fname.disk_name.name = (unsigned char *)name;
3367 fname.disk_name.len = namelen;
3368 } else if (name && namelen) {
3369 struct qstr dname = QSTR_INIT(name, namelen);
3371 struct lu_fid *pfid = NULL;
3375 if (!S_ISDIR(i1->i_mode) && i2 && S_ISDIR(i2->i_mode)) {
3376 /* special case when called from ll_link() */
3381 lookup = (int)(opc == LUSTRE_OPC_ANY);
3383 if (opc == LUSTRE_OPC_ANY && lookup)
3385 rc = ll_setup_filename(dir, &dname, lookup, &fname, pfid);
3387 ll_finish_md_op_data(op_data);
3390 if (pfid && !fid_is_zero(pfid)) {
3392 op_data->op_fid2 = fid;
3393 op_data->op_bias = MDS_FID_OP;
3395 if (fname.disk_name.name &&
3396 fname.disk_name.name != (unsigned char *)name) {
3397 /* op_data->op_name must be freed after use */
3398 op_data->op_flags |= MF_OPNAME_KMALLOCED;
3402 /* In fact LUSTRE_OPC_LOOKUP, LUSTRE_OPC_OPEN
3403 * are LUSTRE_OPC_ANY
3405 if (opc == LUSTRE_OPC_LOOKUP || opc == LUSTRE_OPC_OPEN)
3406 op_data->op_code = LUSTRE_OPC_ANY;
3408 op_data->op_code = opc;
3409 op_data->op_name = fname.disk_name.name;
3410 op_data->op_namelen = fname.disk_name.len;
3411 op_data->op_mode = mode;
3412 op_data->op_mod_time = ktime_get_real_seconds();
3413 op_data->op_fsuid = from_kuid(&init_user_ns, current_fsuid());
3414 op_data->op_fsgid = from_kgid(&init_user_ns, current_fsgid());
3415 op_data->op_cap = current_cap();
3416 op_data->op_mds = 0;
3417 if ((opc == LUSTRE_OPC_CREATE) && (name != NULL) &&
3418 filename_is_volatile(name, namelen, &op_data->op_mds)) {
3419 op_data->op_bias |= MDS_CREATE_VOLATILE;
3421 op_data->op_data = data;
3426 void ll_finish_md_op_data(struct md_op_data *op_data)
3428 ll_unlock_md_op_lsm(op_data);
3429 ll_security_release_secctx(op_data->op_file_secctx,
3430 op_data->op_file_secctx_size);
3431 if (op_data->op_flags & MF_OPNAME_KMALLOCED)
3432 /* allocated via ll_setup_filename called
3433 * from ll_prep_md_op_data
3435 kfree(op_data->op_name);
3436 llcrypt_free_ctx(op_data->op_file_encctx, op_data->op_file_encctx_size);
3437 OBD_FREE_PTR(op_data);
3440 int ll_show_options(struct seq_file *seq, struct dentry *dentry)
3442 struct ll_sb_info *sbi;
3445 LASSERT(seq && dentry);
3446 sbi = ll_s2sbi(dentry->d_sb);
3448 if (test_bit(LL_SBI_NOLCK, sbi->ll_flags))
3449 seq_puts(seq, "nolock");
3451 for (i = 1; ll_sbi_flags_name[i].token != LL_SBI_NUM_MOUNT_OPT; i++) {
3452 /* match_table in some cases has patterns for both enabled and
3453 * disabled cases. Ignore 'no'xxx versions if bit is set.
3455 if (test_bit(ll_sbi_flags_name[i].token, sbi->ll_flags) &&
3456 strncmp(ll_sbi_flags_name[i].pattern, "no", 2)) {
3457 if (ll_sbi_flags_name[i].token ==
3458 LL_SBI_FOREIGN_SYMLINK) {
3459 seq_show_option(seq, "foreign_symlink",
3460 sbi->ll_foreign_symlink_prefix);
3462 seq_printf(seq, ",%s",
3463 ll_sbi_flags_name[i].pattern);
3466 /* You can have either localflock or flock but not
3467 * both. If localflock is set don't print flock or
3470 if (ll_sbi_flags_name[i].token == LL_SBI_LOCALFLOCK)
3472 } else if (!test_bit(ll_sbi_flags_name[i].token, sbi->ll_flags) &&
3473 !strncmp(ll_sbi_flags_name[i].pattern, "no", 2)) {
3474 seq_printf(seq, ",%s",
3475 ll_sbi_flags_name[i].pattern);
3479 llcrypt_show_test_dummy_encryption(seq, ',', dentry->d_sb);
3485 * Get obd name by cmd, and copy out to user space
3487 int ll_get_obd_name(struct inode *inode, unsigned int cmd, unsigned long arg)
3489 struct ll_sb_info *sbi = ll_i2sbi(inode);
3490 struct obd_device *obd;
3493 if (cmd == OBD_IOC_GETNAME_OLD || cmd == OBD_IOC_GETDTNAME)
3494 obd = class_exp2obd(sbi->ll_dt_exp);
3495 else if (cmd == OBD_IOC_GETMDNAME)
3496 obd = class_exp2obd(sbi->ll_md_exp);
3503 if (copy_to_user((void __user *)arg, obd->obd_name,
3504 strlen(obd->obd_name) + 1))
3511 struct work_struct db_work;
3512 struct dentry *db_dentry;
3513 /* Let's hope the path is not too long, 32 bytes for the work struct
3516 char buf[PAGE_SIZE - sizeof(struct work_struct) - sizeof(void *)];
3519 static void ll_dput_later(struct work_struct *work)
3521 struct dname_buf *db = container_of(work, struct dname_buf, db_work);
3523 dput(db->db_dentry);
3524 free_page((unsigned long)db);
3527 static char* ll_d_path(struct dentry *dentry, char *buf, int bufsize)
3534 p.mnt = current->fs->root.mnt;
3536 path = d_path(&p, buf, bufsize);
3541 void ll_dirty_page_discard_warn(struct inode *inode, int ioret)
3543 struct dname_buf *db;
3545 struct dentry *dentry = NULL;
3547 /* this can be called inside spin lock so use GFP_ATOMIC. */
3548 db = (struct dname_buf *)__get_free_page(GFP_ATOMIC);
3551 dentry = d_find_alias(inode);
3553 path = ll_d_path(dentry, db->buf, sizeof(db->buf));
3556 /* The below message is checked in recovery-small.sh test_24b */
3558 "%s: dirty page discard: %s/fid: "DFID"/%s may get corrupted "
3559 "(rc %d)\n", ll_i2sbi(inode)->ll_fsname,
3560 s2lsi(inode->i_sb)->lsi_lmd->lmd_dev,
3561 PFID(ll_inode2fid(inode)),
3562 (path && !IS_ERR(path)) ? path : "", ioret);
3564 if (dentry != NULL) {
3565 /* We cannot dput here since if we happen to be the last holder
3566 * then we can end up waiting for page evictions that
3567 * in turn wait for RPCs that need this instance of ptlrpcd
3568 * (callng brw_interpret->*page_completion*->vmpage_error->here)
3571 INIT_WORK(&db->db_work, ll_dput_later);
3572 db->db_dentry = dentry;
3573 schedule_work(&db->db_work);
3576 free_page((unsigned long)db);
3580 ssize_t ll_copy_user_md(const struct lov_user_md __user *md,
3581 struct lov_user_md **kbuf)
3583 struct lov_user_md lum;
3587 if (copy_from_user(&lum, md, sizeof(lum)))
3590 lum_size = ll_lov_user_md_size(&lum);
3594 OBD_ALLOC_LARGE(*kbuf, lum_size);
3598 if (copy_from_user(*kbuf, md, lum_size) != 0) {
3599 OBD_FREE_LARGE(*kbuf, lum_size);
3607 * Compute llite root squash state after a change of root squash
3608 * configuration setting or add/remove of a lnet nid
3610 void ll_compute_rootsquash_state(struct ll_sb_info *sbi)
3612 struct root_squash_info *squash = &sbi->ll_squash;
3615 struct lnet_processid id;
3617 /* Update norootsquash flag */
3618 spin_lock(&squash->rsi_lock);
3619 if (list_empty(&squash->rsi_nosquash_nids))
3620 clear_bit(LL_SBI_NOROOTSQUASH, sbi->ll_flags);
3622 /* Do not apply root squash as soon as one of our NIDs is
3623 * in the nosquash_nids list */
3626 while (LNetGetId(i++, &id) != -ENOENT) {
3627 if (nid_is_lo0(&id.nid))
3629 if (cfs_match_nid(lnet_nid_to_nid4(&id.nid),
3630 &squash->rsi_nosquash_nids)) {
3636 set_bit(LL_SBI_NOROOTSQUASH, sbi->ll_flags);
3638 clear_bit(LL_SBI_NOROOTSQUASH, sbi->ll_flags);
3640 spin_unlock(&squash->rsi_lock);
3644 * Parse linkea content to extract information about a given hardlink
3646 * \param[in] ldata - Initialized linkea data
3647 * \param[in] linkno - Link identifier
3648 * \param[out] parent_fid - The entry's parent FID
3649 * \param[out] ln - Entry name destination buffer
3651 * \retval 0 on success
3652 * \retval Appropriate negative error code on failure
3654 static int ll_linkea_decode(struct linkea_data *ldata, unsigned int linkno,
3655 struct lu_fid *parent_fid, struct lu_name *ln)
3661 rc = linkea_init_with_rec(ldata);
3665 if (linkno >= ldata->ld_leh->leh_reccount)
3666 /* beyond last link */
3669 linkea_first_entry(ldata);
3670 for (idx = 0; ldata->ld_lee != NULL; idx++) {
3671 linkea_entry_unpack(ldata->ld_lee, &ldata->ld_reclen, ln,
3676 linkea_next_entry(ldata);
3686 * Get parent FID and name of an identified link. Operation is performed for
3687 * a given link number, letting the caller iterate over linkno to list one or
3688 * all links of an entry.
3690 * \param[in] file - File descriptor against which to perform the operation
3691 * \param[in,out] arg - User-filled structure containing the linkno to operate
3692 * on and the available size. It is eventually filled with
3693 * the requested information or left untouched on error
3695 * \retval - 0 on success
3696 * \retval - Appropriate negative error code on failure
3698 int ll_getparent(struct file *file, struct getparent __user *arg)
3700 struct inode *inode = file_inode(file);
3701 struct linkea_data *ldata;
3702 struct lu_buf buf = LU_BUF_NULL;
3704 struct lu_fid parent_fid;
3711 if (!capable(CAP_DAC_READ_SEARCH) &&
3712 !test_bit(LL_SBI_USER_FID2PATH, ll_i2sbi(inode)->ll_flags))
3715 if (get_user(name_size, &arg->gp_name_size))
3718 if (get_user(linkno, &arg->gp_linkno))
3721 if (name_size > PATH_MAX)
3724 OBD_ALLOC(ldata, sizeof(*ldata));
3728 rc = linkea_data_new(ldata, &buf);
3730 GOTO(ldata_free, rc);
3732 rc = ll_xattr_list(inode, XATTR_NAME_LINK, XATTR_TRUSTED_T, buf.lb_buf,
3733 buf.lb_len, OBD_MD_FLXATTR);
3737 rc = ll_linkea_decode(ldata, linkno, &parent_fid, &ln);
3741 if (ln.ln_namelen >= name_size)
3742 GOTO(lb_free, rc = -EOVERFLOW);
3744 if (copy_to_user(&arg->gp_fid, &parent_fid, sizeof(arg->gp_fid)))
3745 GOTO(lb_free, rc = -EFAULT);
3747 if (copy_to_user(&arg->gp_name, ln.ln_name, ln.ln_namelen))
3748 GOTO(lb_free, rc = -EFAULT);
3750 if (put_user('\0', arg->gp_name + ln.ln_namelen))
3751 GOTO(lb_free, rc = -EFAULT);
3756 OBD_FREE(ldata, sizeof(*ldata));