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/types.h>
44 #include <libcfs/linux/linux-uuid.h>
45 #include <linux/version.h>
47 #include <linux/user_namespace.h>
48 #include <linux/delay.h>
49 #include <linux/uidgid.h>
50 #include <linux/fs_struct.h>
52 #ifndef HAVE_CPUS_READ_LOCK
53 #include <libcfs/linux/linux-cpu.h>
55 #include <libcfs/linux/linux-misc.h>
56 #include <uapi/linux/lustre/lustre_ioctl.h>
57 #ifdef HAVE_UAPI_LINUX_MOUNT_H
58 #include <uapi/linux/mount.h>
61 #include <lustre_ha.h>
62 #include <lustre_dlm.h>
63 #include <lprocfs_status.h>
64 #include <lustre_disk.h>
65 #include <uapi/linux/lustre/lustre_param.h>
66 #include <lustre_log.h>
67 #include <cl_object.h>
68 #include <obd_cksum.h>
69 #include "llite_internal.h"
71 struct kmem_cache *ll_file_data_slab;
74 #define log2(n) ffz(~(n))
78 * If there is only one number of core visible to Lustre,
79 * async readahead will be disabled, to avoid massive over
80 * subscription, we use 1/2 of active cores as default max
81 * async readahead requests.
83 static inline unsigned int ll_get_ra_async_max_active(void)
85 return cfs_cpt_weight(cfs_cpt_tab, CFS_CPT_ANY) >> 1;
88 static struct ll_sb_info *ll_init_sbi(void)
90 struct ll_sb_info *sbi = NULL;
92 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 for (i = 0; i <= LL_PROCESS_HIST_MAX; i++) {
175 spin_lock_init(&sbi->ll_rw_extents_info.pp_extents[i].
177 spin_lock_init(&sbi->ll_rw_extents_info.pp_extents[i].
181 /* metadata statahead is enabled by default */
182 sbi->ll_sa_running_max = LL_SA_RUNNING_DEF;
183 sbi->ll_sa_max = LL_SA_RPC_DEF;
184 atomic_set(&sbi->ll_sa_total, 0);
185 atomic_set(&sbi->ll_sa_wrong, 0);
186 atomic_set(&sbi->ll_sa_running, 0);
187 atomic_set(&sbi->ll_agl_total, 0);
188 set_bit(LL_SBI_AGL_ENABLED, sbi->ll_flags);
189 set_bit(LL_SBI_FAST_READ, sbi->ll_flags);
190 set_bit(LL_SBI_TINY_WRITE, sbi->ll_flags);
191 set_bit(LL_SBI_PARALLEL_DIO, sbi->ll_flags);
192 ll_sbi_set_encrypt(sbi, true);
195 sbi->ll_squash.rsi_uid = 0;
196 sbi->ll_squash.rsi_gid = 0;
197 INIT_LIST_HEAD(&sbi->ll_squash.rsi_nosquash_nids);
198 spin_lock_init(&sbi->ll_squash.rsi_lock);
200 /* Per-filesystem file heat */
201 sbi->ll_heat_decay_weight = SBI_DEFAULT_HEAT_DECAY_WEIGHT;
202 sbi->ll_heat_period_second = SBI_DEFAULT_HEAT_PERIOD_SECOND;
204 /* Per-fs open heat level before requesting open lock */
205 sbi->ll_oc_thrsh_count = SBI_DEFAULT_OPENCACHE_THRESHOLD_COUNT;
206 sbi->ll_oc_max_ms = SBI_DEFAULT_OPENCACHE_THRESHOLD_MAX_MS;
207 sbi->ll_oc_thrsh_ms = SBI_DEFAULT_OPENCACHE_THRESHOLD_MS;
210 if (sbi->ll_foreign_symlink_prefix)
211 OBD_FREE(sbi->ll_foreign_symlink_prefix, sizeof("/mnt/"));
213 cl_cache_decref(sbi->ll_cache);
214 sbi->ll_cache = NULL;
216 destroy_workqueue(sbi->ll_ra_info.ll_readahead_wq);
218 pcc_super_fini(&sbi->ll_pcc_super);
224 static void ll_free_sbi(struct super_block *sb)
226 struct ll_sb_info *sbi = ll_s2sbi(sb);
230 if (!list_empty(&sbi->ll_squash.rsi_nosquash_nids))
231 cfs_free_nidlist(&sbi->ll_squash.rsi_nosquash_nids);
232 if (sbi->ll_ra_info.ll_readahead_wq)
233 destroy_workqueue(sbi->ll_ra_info.ll_readahead_wq);
234 if (sbi->ll_cache != NULL) {
235 cl_cache_decref(sbi->ll_cache);
236 sbi->ll_cache = NULL;
238 if (sbi->ll_foreign_symlink_prefix) {
239 OBD_FREE(sbi->ll_foreign_symlink_prefix,
240 sbi->ll_foreign_symlink_prefix_size);
241 sbi->ll_foreign_symlink_prefix = NULL;
243 if (sbi->ll_foreign_symlink_upcall) {
244 OBD_FREE(sbi->ll_foreign_symlink_upcall,
245 strlen(sbi->ll_foreign_symlink_upcall) +
247 sbi->ll_foreign_symlink_upcall = NULL;
249 if (sbi->ll_foreign_symlink_upcall_items) {
251 int nb_items = sbi->ll_foreign_symlink_upcall_nb_items;
252 struct ll_foreign_symlink_upcall_item *items =
253 sbi->ll_foreign_symlink_upcall_items;
255 for (i = 0 ; i < nb_items; i++)
256 if (items[i].type == STRING_TYPE)
257 OBD_FREE(items[i].string,
260 OBD_FREE_LARGE(items, nb_items *
261 sizeof(struct ll_foreign_symlink_upcall_item));
262 sbi->ll_foreign_symlink_upcall_items = NULL;
264 pcc_super_fini(&sbi->ll_pcc_super);
265 OBD_FREE(sbi, sizeof(*sbi));
270 static int client_common_fill_super(struct super_block *sb, char *md, char *dt)
272 struct inode *root = NULL;
273 struct ll_sb_info *sbi = ll_s2sbi(sb);
274 struct obd_statfs *osfs = NULL;
275 struct ptlrpc_request *request = NULL;
276 struct obd_connect_data *data = NULL;
277 struct obd_uuid *uuid;
278 struct md_op_data *op_data;
279 struct lustre_md lmd;
281 int size, err, checksum;
285 sbi->ll_md_obd = class_name2obd(md);
286 if (!sbi->ll_md_obd) {
287 CERROR("MD %s: not setup or attached\n", md);
301 /* pass client page size via ocd_grant_blkbits, the server should report
302 * back its backend blocksize for grant calculation purpose */
303 data->ocd_grant_blkbits = PAGE_SHIFT;
305 /* indicate MDT features supported by this client */
306 data->ocd_connect_flags = OBD_CONNECT_IBITS | OBD_CONNECT_NODEVOH |
307 OBD_CONNECT_ATTRFID | OBD_CONNECT_GRANT |
308 OBD_CONNECT_VERSION | OBD_CONNECT_BRW_SIZE |
309 OBD_CONNECT_SRVLOCK |
310 OBD_CONNECT_MDS_CAPA | OBD_CONNECT_OSS_CAPA |
311 OBD_CONNECT_CANCELSET | OBD_CONNECT_FID |
312 OBD_CONNECT_AT | OBD_CONNECT_LOV_V3 |
313 OBD_CONNECT_VBR | OBD_CONNECT_FULL20 |
314 OBD_CONNECT_64BITHASH |
315 OBD_CONNECT_EINPROGRESS |
316 OBD_CONNECT_JOBSTATS | OBD_CONNECT_LVB_TYPE |
317 OBD_CONNECT_LAYOUTLOCK | OBD_CONNECT_PINGLESS|
318 OBD_CONNECT_MAX_EASIZE |
319 OBD_CONNECT_FLOCK_DEAD |
320 OBD_CONNECT_DISP_STRIPE | OBD_CONNECT_LFSCK |
321 OBD_CONNECT_OPEN_BY_FID |
322 OBD_CONNECT_DIR_STRIPE |
323 OBD_CONNECT_BULK_MBITS | OBD_CONNECT_CKSUM |
324 OBD_CONNECT_SUBTREE |
325 OBD_CONNECT_MULTIMODRPCS |
326 OBD_CONNECT_GRANT_PARAM |
327 OBD_CONNECT_SHORTIO | OBD_CONNECT_FLAGS2;
329 data->ocd_connect_flags2 = OBD_CONNECT2_DIR_MIGRATE |
330 OBD_CONNECT2_SUM_STATFS |
331 OBD_CONNECT2_OVERSTRIPING |
333 OBD_CONNECT2_LOCK_CONVERT |
334 OBD_CONNECT2_ARCHIVE_ID_ARRAY |
335 OBD_CONNECT2_INC_XID |
337 OBD_CONNECT2_ASYNC_DISCARD |
339 OBD_CONNECT2_CRUSH | OBD_CONNECT2_LSEEK |
340 OBD_CONNECT2_GETATTR_PFID |
341 OBD_CONNECT2_DOM_LVB |
342 OBD_CONNECT2_REP_MBITS |
343 OBD_CONNECT2_ATOMIC_OPEN_LOCK;
345 #ifdef HAVE_LRU_RESIZE_SUPPORT
346 if (test_bit(LL_SBI_LRU_RESIZE, sbi->ll_flags))
347 data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
349 data->ocd_connect_flags |= OBD_CONNECT_ACL_FLAGS;
351 data->ocd_cksum_types = obd_cksum_types_supported_client();
353 if (OBD_FAIL_CHECK(OBD_FAIL_MDC_LIGHTWEIGHT))
354 /* flag mdc connection as lightweight, only used for test
355 * purpose, use with care */
356 data->ocd_connect_flags |= OBD_CONNECT_LIGHTWEIGHT;
358 data->ocd_ibits_known = MDS_INODELOCK_FULL;
359 data->ocd_version = LUSTRE_VERSION_CODE;
361 if (sb->s_flags & SB_RDONLY)
362 data->ocd_connect_flags |= OBD_CONNECT_RDONLY;
363 if (test_bit(LL_SBI_USER_XATTR, sbi->ll_flags))
364 data->ocd_connect_flags |= OBD_CONNECT_XATTR;
367 /* Setting this indicates we correctly support S_NOSEC (See kernel
368 * commit 9e1f1de02c2275d7172e18dc4e7c2065777611bf)
370 sb->s_flags |= SB_NOSEC;
372 sbi->ll_fop = ll_select_file_operations(sbi);
374 /* always ping even if server suppress_pings */
375 if (test_bit(LL_SBI_ALWAYS_PING, sbi->ll_flags))
376 data->ocd_connect_flags &= ~OBD_CONNECT_PINGLESS;
378 obd_connect_set_secctx(data);
379 if (ll_sbi_has_encrypt(sbi))
380 obd_connect_set_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;
461 if (test_bit(LL_SBI_USER_XATTR, sbi->ll_flags) &&
462 !(data->ocd_connect_flags & OBD_CONNECT_XATTR)) {
463 LCONSOLE_INFO("Disabling user_xattr feature because "
464 "it is not supported on the server\n");
465 clear_bit(LL_SBI_USER_XATTR, sbi->ll_flags);
468 if (data->ocd_connect_flags & OBD_CONNECT_ACL) {
470 sb->s_flags |= SB_POSIXACL;
472 set_bit(LL_SBI_ACL, sbi->ll_flags);
474 LCONSOLE_INFO("client wants to enable acl, but mdt not!\n");
476 sb->s_flags &= ~SB_POSIXACL;
478 clear_bit(LL_SBI_ACL, sbi->ll_flags);
481 if (data->ocd_connect_flags & OBD_CONNECT_64BITHASH)
482 set_bit(LL_SBI_64BIT_HASH, sbi->ll_flags);
484 if (data->ocd_connect_flags & OBD_CONNECT_LAYOUTLOCK)
485 set_bit(LL_SBI_LAYOUT_LOCK, sbi->ll_flags);
487 if (obd_connect_has_secctx(data))
488 set_bit(LL_SBI_FILE_SECCTX, sbi->ll_flags);
490 if (ll_sbi_has_encrypt(sbi) && !obd_connect_has_enc(data)) {
491 if (ll_sbi_has_test_dummy_encryption(sbi))
492 LCONSOLE_WARN("%s: server %s does not support encryption feature, encryption deactivated.\n",
494 sbi->ll_md_exp->exp_obd->obd_name);
495 ll_sbi_set_encrypt(sbi, false);
498 if (data->ocd_ibits_known & MDS_INODELOCK_XATTR) {
499 if (!(data->ocd_connect_flags & OBD_CONNECT_MAX_EASIZE)) {
500 LCONSOLE_INFO("%s: disabling xattr cache due to "
501 "unknown maximum xattr size.\n", dt);
502 } else if (!sbi->ll_xattr_cache_set) {
503 /* If xattr_cache is already set (no matter 0 or 1)
504 * during processing llog, it won't be enabled here. */
505 set_bit(LL_SBI_XATTR_CACHE, sbi->ll_flags);
506 sbi->ll_xattr_cache_enabled = 1;
510 sbi->ll_dt_obd = class_name2obd(dt);
511 if (!sbi->ll_dt_obd) {
512 CERROR("DT %s: not setup or attached\n", dt);
513 GOTO(out_md_fid, err = -ENODEV);
516 /* pass client page size via ocd_grant_blkbits, the server should report
517 * back its backend blocksize for grant calculation purpose */
518 data->ocd_grant_blkbits = PAGE_SHIFT;
520 /* indicate OST features supported by this client */
521 data->ocd_connect_flags = OBD_CONNECT_GRANT | OBD_CONNECT_VERSION |
522 OBD_CONNECT_REQPORTAL | OBD_CONNECT_BRW_SIZE |
523 OBD_CONNECT_CANCELSET | OBD_CONNECT_FID |
524 OBD_CONNECT_SRVLOCK |
525 OBD_CONNECT_AT | OBD_CONNECT_OSS_CAPA |
526 OBD_CONNECT_VBR | OBD_CONNECT_FULL20 |
527 OBD_CONNECT_64BITHASH | OBD_CONNECT_MAXBYTES |
528 OBD_CONNECT_EINPROGRESS |
529 OBD_CONNECT_JOBSTATS | OBD_CONNECT_LVB_TYPE |
530 OBD_CONNECT_LAYOUTLOCK |
531 OBD_CONNECT_PINGLESS | OBD_CONNECT_LFSCK |
532 OBD_CONNECT_BULK_MBITS | OBD_CONNECT_SHORTIO |
533 OBD_CONNECT_FLAGS2 | OBD_CONNECT_GRANT_SHRINK;
534 data->ocd_connect_flags2 = OBD_CONNECT2_LOCKAHEAD |
535 OBD_CONNECT2_INC_XID | OBD_CONNECT2_LSEEK |
536 OBD_CONNECT2_REP_MBITS;
538 if (!OBD_FAIL_CHECK(OBD_FAIL_OSC_CONNECT_GRANT_PARAM))
539 data->ocd_connect_flags |= OBD_CONNECT_GRANT_PARAM;
541 /* OBD_CONNECT_CKSUM should always be set, even if checksums are
542 * disabled by default, because it can still be enabled on the
543 * fly via /sys. As a consequence, we still need to come to an
544 * agreement on the supported algorithms at connect time
546 data->ocd_connect_flags |= OBD_CONNECT_CKSUM;
548 if (OBD_FAIL_CHECK(OBD_FAIL_OSC_CKSUM_ADLER_ONLY))
549 data->ocd_cksum_types = OBD_CKSUM_ADLER;
551 data->ocd_cksum_types = obd_cksum_types_supported_client();
553 #ifdef HAVE_LRU_RESIZE_SUPPORT
554 data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
556 /* always ping even if server suppress_pings */
557 if (test_bit(LL_SBI_ALWAYS_PING, sbi->ll_flags))
558 data->ocd_connect_flags &= ~OBD_CONNECT_PINGLESS;
560 if (ll_sbi_has_encrypt(sbi))
561 obd_connect_set_enc(data);
563 CDEBUG(D_RPCTRACE, "ocd_connect_flags: %#llx ocd_version: %d "
564 "ocd_grant: %d\n", data->ocd_connect_flags,
565 data->ocd_version, data->ocd_grant);
567 sbi->ll_dt_obd->obd_upcall.onu_owner = &sbi->ll_lco;
568 sbi->ll_dt_obd->obd_upcall.onu_upcall = cl_ocd_update;
570 data->ocd_brw_size = DT_MAX_BRW_SIZE;
572 err = obd_connect(NULL, &sbi->ll_dt_exp, sbi->ll_dt_obd,
573 &sbi->ll_sb_uuid, data, sbi->ll_cache);
575 LCONSOLE_ERROR_MSG(0x150, "An OST (dt %s) is performing "
576 "recovery, of which this client is not a "
577 "part. Please wait for recovery to "
578 "complete, abort, or time out.\n", dt);
581 CERROR("%s: Cannot connect to %s: rc = %d\n",
582 sbi->ll_dt_exp->exp_obd->obd_name, dt, err);
586 if (ll_sbi_has_encrypt(sbi) &&
587 !obd_connect_has_enc(&sbi->ll_dt_obd->u.lov.lov_ocd)) {
588 if (ll_sbi_has_test_dummy_encryption(sbi))
589 LCONSOLE_WARN("%s: server %s does not support encryption feature, encryption deactivated.\n",
591 ll_sbi_set_encrypt(sbi, false);
592 } else if (ll_sbi_has_test_dummy_encryption(sbi)) {
593 LCONSOLE_WARN("Test dummy encryption mode enabled\n");
596 sbi->ll_dt_exp->exp_connect_data = *data;
598 /* Don't change value if it was specified in the config log */
599 if (sbi->ll_ra_info.ra_max_read_ahead_whole_pages == -1) {
600 sbi->ll_ra_info.ra_max_read_ahead_whole_pages =
601 max_t(unsigned long, SBI_DEFAULT_READ_AHEAD_WHOLE_MAX,
602 (data->ocd_brw_size >> PAGE_SHIFT));
603 if (sbi->ll_ra_info.ra_max_read_ahead_whole_pages >
604 sbi->ll_ra_info.ra_max_pages_per_file)
605 sbi->ll_ra_info.ra_max_read_ahead_whole_pages =
606 sbi->ll_ra_info.ra_max_pages_per_file;
609 err = obd_fid_init(sbi->ll_dt_exp->exp_obd, sbi->ll_dt_exp,
610 LUSTRE_SEQ_METADATA);
612 CERROR("%s: Can't init data layer FID infrastructure, "
613 "rc = %d\n", sbi->ll_dt_exp->exp_obd->obd_name, err);
617 mutex_lock(&sbi->ll_lco.lco_lock);
618 sbi->ll_lco.lco_flags = data->ocd_connect_flags;
619 sbi->ll_lco.lco_md_exp = sbi->ll_md_exp;
620 sbi->ll_lco.lco_dt_exp = sbi->ll_dt_exp;
621 mutex_unlock(&sbi->ll_lco.lco_lock);
623 fid_zero(&sbi->ll_root_fid);
624 err = md_get_root(sbi->ll_md_exp, get_mount_fileset(sb),
627 CERROR("cannot mds_connect: rc = %d\n", err);
628 GOTO(out_lock_cn_cb, err);
630 if (!fid_is_sane(&sbi->ll_root_fid)) {
631 CERROR("%s: Invalid root fid "DFID" during mount\n",
632 sbi->ll_md_exp->exp_obd->obd_name,
633 PFID(&sbi->ll_root_fid));
634 GOTO(out_lock_cn_cb, err = -EINVAL);
636 CDEBUG(D_SUPER, "rootfid "DFID"\n", PFID(&sbi->ll_root_fid));
638 sb->s_op = &lustre_super_operations;
639 sb->s_xattr = ll_xattr_handlers;
640 #if THREAD_SIZE >= 8192 /*b=17630*/
641 sb->s_export_op = &lustre_export_operations;
643 #ifdef HAVE_LUSTRE_CRYPTO
644 llcrypt_set_ops(sb, &lustre_cryptops);
648 * XXX: move this to after cbd setup? */
649 valid = OBD_MD_FLGETATTR | OBD_MD_FLBLOCKS | OBD_MD_FLMODEASIZE;
650 if (test_bit(LL_SBI_ACL, sbi->ll_flags))
651 valid |= OBD_MD_FLACL;
653 OBD_ALLOC_PTR(op_data);
655 GOTO(out_lock_cn_cb, err = -ENOMEM);
657 op_data->op_fid1 = sbi->ll_root_fid;
658 op_data->op_mode = 0;
659 op_data->op_valid = valid;
661 err = md_getattr(sbi->ll_md_exp, op_data, &request);
663 OBD_FREE_PTR(op_data);
665 CERROR("%s: md_getattr failed for root: rc = %d\n",
666 sbi->ll_md_exp->exp_obd->obd_name, err);
667 GOTO(out_lock_cn_cb, err);
670 err = md_get_lustre_md(sbi->ll_md_exp, &request->rq_pill,
671 sbi->ll_dt_exp, sbi->ll_md_exp, &lmd);
673 CERROR("failed to understand root inode md: rc = %d\n", err);
674 ptlrpc_req_finished(request);
675 GOTO(out_lock_cn_cb, err);
678 LASSERT(fid_is_sane(&sbi->ll_root_fid));
679 api32 = test_bit(LL_SBI_32BIT_API, sbi->ll_flags);
680 root = ll_iget(sb, cl_fid_build_ino(&sbi->ll_root_fid, api32), &lmd);
681 md_free_lustre_md(sbi->ll_md_exp, &lmd);
682 ptlrpc_req_finished(request);
686 err = IS_ERR(root) ? PTR_ERR(root) : -EBADF;
688 CERROR("%s: bad ll_iget() for root: rc = %d\n",
689 sbi->ll_fsname, err);
693 checksum = test_bit(LL_SBI_CHECKSUM, sbi->ll_flags);
694 if (sbi->ll_checksum_set) {
695 err = obd_set_info_async(NULL, sbi->ll_dt_exp,
696 sizeof(KEY_CHECKSUM), KEY_CHECKSUM,
697 sizeof(checksum), &checksum, NULL);
699 CERROR("%s: Set checksum failed: rc = %d\n",
700 sbi->ll_dt_exp->exp_obd->obd_name, err);
706 sb->s_root = d_make_root(root);
707 if (sb->s_root == NULL) {
709 CERROR("%s: can't make root dentry: rc = %d\n",
710 sbi->ll_fsname, err);
714 sbi->ll_sdev_orig = sb->s_dev;
716 /* We set sb->s_dev equal on all lustre clients in order to support
717 * NFS export clustering. NFSD requires that the FSID be the same
719 /* s_dev is also used in lt_compare() to compare two fs, but that is
720 * only a node-local comparison. */
721 uuid = obd_get_uuid(sbi->ll_md_exp);
723 sb->s_dev = get_uuid2int(uuid->uuid, strlen(uuid->uuid));
730 if (sbi->ll_dt_obd) {
731 err = sysfs_create_link(&sbi->ll_kset.kobj,
732 &sbi->ll_dt_obd->obd_kset.kobj,
733 sbi->ll_dt_obd->obd_type->typ_name);
735 CERROR("%s: could not register %s in llite: rc = %d\n",
736 dt, sbi->ll_fsname, err);
741 if (sbi->ll_md_obd) {
742 err = sysfs_create_link(&sbi->ll_kset.kobj,
743 &sbi->ll_md_obd->obd_kset.kobj,
744 sbi->ll_md_obd->obd_type->typ_name);
746 CERROR("%s: could not register %s in llite: rc = %d\n",
747 md, sbi->ll_fsname, err);
756 obd_fid_fini(sbi->ll_dt_exp->exp_obd);
758 obd_disconnect(sbi->ll_dt_exp);
759 sbi->ll_dt_exp = NULL;
760 sbi->ll_dt_obd = NULL;
762 obd_fid_fini(sbi->ll_md_exp->exp_obd);
764 obd_disconnect(sbi->ll_md_exp);
765 sbi->ll_md_exp = NULL;
766 sbi->ll_md_obd = NULL;
775 int ll_get_max_mdsize(struct ll_sb_info *sbi, int *lmmsize)
779 size = sizeof(*lmmsize);
780 rc = obd_get_info(NULL, sbi->ll_dt_exp, sizeof(KEY_MAX_EASIZE),
781 KEY_MAX_EASIZE, &size, lmmsize);
783 CERROR("%s: cannot get max LOV EA size: rc = %d\n",
784 sbi->ll_dt_exp->exp_obd->obd_name, rc);
788 CDEBUG(D_INFO, "max LOV ea size: %d\n", *lmmsize);
791 rc = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_MAX_EASIZE),
792 KEY_MAX_EASIZE, &size, lmmsize);
794 CERROR("Get max mdsize error rc %d\n", rc);
796 CDEBUG(D_INFO, "max LMV ea size: %d\n", *lmmsize);
802 * Get the value of the default_easize parameter.
804 * \see client_obd::cl_default_mds_easize
806 * \param[in] sbi superblock info for this filesystem
807 * \param[out] lmmsize pointer to storage location for value
809 * \retval 0 on success
810 * \retval negative negated errno on failure
812 int ll_get_default_mdsize(struct ll_sb_info *sbi, int *lmmsize)
817 rc = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_DEFAULT_EASIZE),
818 KEY_DEFAULT_EASIZE, &size, lmmsize);
820 CERROR("Get default mdsize error rc %d\n", rc);
826 * Set the default_easize parameter to the given value.
828 * \see client_obd::cl_default_mds_easize
830 * \param[in] sbi superblock info for this filesystem
831 * \param[in] lmmsize the size to set
833 * \retval 0 on success
834 * \retval negative negated errno on failure
836 int ll_set_default_mdsize(struct ll_sb_info *sbi, int lmmsize)
840 if (lmmsize < sizeof(struct lov_mds_md) ||
841 lmmsize > OBD_MAX_DEFAULT_EA_SIZE)
844 rc = obd_set_info_async(NULL, sbi->ll_md_exp,
845 sizeof(KEY_DEFAULT_EASIZE), KEY_DEFAULT_EASIZE,
846 sizeof(int), &lmmsize, NULL);
851 static void client_common_put_super(struct super_block *sb)
853 struct ll_sb_info *sbi = ll_s2sbi(sb);
858 obd_fid_fini(sbi->ll_dt_exp->exp_obd);
859 obd_disconnect(sbi->ll_dt_exp);
860 sbi->ll_dt_exp = NULL;
862 ll_debugfs_unregister_super(sb);
864 obd_fid_fini(sbi->ll_md_exp->exp_obd);
865 obd_disconnect(sbi->ll_md_exp);
866 sbi->ll_md_exp = NULL;
871 void ll_kill_super(struct super_block *sb)
873 struct ll_sb_info *sbi;
877 if (!(sb->s_flags & SB_ACTIVE))
881 /* we need restore s_dev from changed for clustred NFS before put_super
882 * because new kernels have cached s_dev and change sb->s_dev in
883 * put_super not affected real removing devices */
885 sb->s_dev = sbi->ll_sdev_orig;
887 /* wait running statahead threads to quit */
888 while (atomic_read(&sbi->ll_sa_running) > 0)
889 schedule_timeout_uninterruptible(
890 cfs_time_seconds(1) >> 3);
896 /* Since we use this table for ll_sbi_flags_seq_show make
897 * sure what you want displayed for a specific token that
898 * is listed more than once below be listed first. For
899 * example we want "checksum" displayed, not "nochecksum"
902 static const match_table_t ll_sbi_flags_name = {
903 {LL_SBI_NOLCK, "nolock"},
904 {LL_SBI_CHECKSUM, "checksum"},
905 {LL_SBI_CHECKSUM, "nochecksum"},
906 {LL_SBI_LOCALFLOCK, "localflock"},
907 {LL_SBI_FLOCK, "flock"},
908 {LL_SBI_FLOCK, "noflock"},
909 {LL_SBI_USER_XATTR, "user_xattr"},
910 {LL_SBI_USER_XATTR, "nouser_xattr"},
911 {LL_SBI_LRU_RESIZE, "lruresize"},
912 {LL_SBI_LRU_RESIZE, "nolruresize"},
913 {LL_SBI_LAZYSTATFS, "lazystatfs"},
914 {LL_SBI_LAZYSTATFS, "nolazystatfs"},
915 {LL_SBI_32BIT_API, "32bitapi"},
916 {LL_SBI_USER_FID2PATH, "user_fid2path"},
917 {LL_SBI_USER_FID2PATH, "nouser_fid2path"},
918 {LL_SBI_VERBOSE, "verbose"},
919 {LL_SBI_VERBOSE, "noverbose"},
920 {LL_SBI_ALWAYS_PING, "always_ping"},
921 {LL_SBI_TEST_DUMMY_ENCRYPTION, "test_dummy_encryption"},
922 {LL_SBI_ENCRYPT, "encrypt"},
923 {LL_SBI_ENCRYPT, "noencrypt"},
924 {LL_SBI_FOREIGN_SYMLINK, "foreign_symlink=%s"},
925 {LL_SBI_NUM_MOUNT_OPT, NULL},
928 {LL_SBI_AGL_ENABLED, "agl"},
929 {LL_SBI_64BIT_HASH, "64bit_hash"},
930 {LL_SBI_LAYOUT_LOCK, "layout"},
931 {LL_SBI_XATTR_CACHE, "xattr_cache"},
932 {LL_SBI_NOROOTSQUASH, "norootsquash"},
933 {LL_SBI_FAST_READ, "fast_read"},
934 {LL_SBI_FILE_SECCTX, "file_secctx"},
935 {LL_SBI_TINY_WRITE, "tiny_write"},
936 {LL_SBI_FILE_HEAT, "file_heat"},
937 {LL_SBI_PARALLEL_DIO, "parallel_dio"},
940 int ll_sbi_flags_seq_show(struct seq_file *m, void *v)
942 struct super_block *sb = m->private;
945 for (i = 0; i < LL_SBI_NUM_FLAGS; i++) {
948 if (!test_bit(i, ll_s2sbi(sb)->ll_flags))
951 for (j = 0; j < ARRAY_SIZE(ll_sbi_flags_name); j++) {
952 if (ll_sbi_flags_name[j].token == i &&
953 ll_sbi_flags_name[j].pattern) {
955 ll_sbi_flags_name[j].pattern);
964 /* non-client-specific mount options are parsed in lmd_parse */
965 static int ll_options(char *options, struct super_block *sb)
967 struct ll_sb_info *sbi = ll_s2sbi(sb);
968 char *s2, *s1, *opts;
974 /* Don't stomp on lmd_opts */
975 opts = kstrdup(options, GFP_KERNEL);
981 CDEBUG(D_CONFIG, "Parsing opts %s\n", options);
983 while ((s1 = strsep(&opts, ",")) != NULL) {
984 substring_t args[MAX_OPT_ARGS];
985 bool turn_off = false;
991 CDEBUG(D_SUPER, "next opt=%s\n", s1);
993 if (strncmp(s1, "no", 2) == 0)
997 * Initialize args struct so we know whether arg was
998 * found; some options take optional arguments.
1001 args[0].from = NULL;
1002 token = match_token(s1, ll_sbi_flags_name, args);
1003 if (token == LL_SBI_NUM_MOUNT_OPT) {
1004 if (match_wildcard("context", s1) ||
1005 match_wildcard("fscontext", s1) ||
1006 match_wildcard("defcontext", s1) ||
1007 match_wildcard("rootcontext",s1))
1010 LCONSOLE_ERROR_MSG(0x152,
1011 "Unknown option '%s', won't mount.\n",
1018 case LL_SBI_32BIT_API:
1019 case LL_SBI_64BIT_HASH:
1020 case LL_SBI_ALWAYS_PING:
1021 set_bit(token, sbi->ll_flags);
1025 clear_bit(LL_SBI_LOCALFLOCK, sbi->ll_flags);
1027 clear_bit(LL_SBI_FLOCK, sbi->ll_flags);
1029 set_bit(token, sbi->ll_flags);
1032 case LL_SBI_LOCALFLOCK:
1033 clear_bit(LL_SBI_FLOCK, sbi->ll_flags);
1034 set_bit(token, sbi->ll_flags);
1037 case LL_SBI_CHECKSUM:
1038 sbi->ll_checksum_set = 1;
1040 case LL_SBI_USER_XATTR:
1041 case LL_SBI_USER_FID2PATH:
1042 case LL_SBI_LRU_RESIZE:
1043 case LL_SBI_LAZYSTATFS:
1044 case LL_SBI_VERBOSE:
1046 clear_bit(token, sbi->ll_flags);
1048 set_bit(token, sbi->ll_flags);
1050 case LL_SBI_TEST_DUMMY_ENCRYPTION: {
1051 #ifdef HAVE_LUSTRE_CRYPTO
1052 set_bit(token, sbi->ll_flags);
1054 LCONSOLE_WARN("Test dummy encryption mount option ignored: encryption not supported\n");
1058 case LL_SBI_ENCRYPT:
1059 #ifdef HAVE_LUSTRE_CRYPTO
1061 clear_bit(token, sbi->ll_flags);
1063 set_bit(token, sbi->ll_flags);
1065 LCONSOLE_WARN("noencrypt or encrypt mount option ignored: encryption not supported\n");
1068 case LL_SBI_FOREIGN_SYMLINK:
1069 /* non-default prefix provided ? */
1074 /* path must be absolute */
1075 if (args->from[0] != '/') {
1076 LCONSOLE_ERROR_MSG(0x152,
1077 "foreign prefix '%s' must be an absolute path\n",
1082 old_len = sbi->ll_foreign_symlink_prefix_size;
1083 old = sbi->ll_foreign_symlink_prefix;
1084 /* alloc for path length and '\0' */
1085 sbi->ll_foreign_symlink_prefix = match_strdup(args);
1086 if (!sbi->ll_foreign_symlink_prefix) {
1087 /* restore previous */
1088 sbi->ll_foreign_symlink_prefix = old;
1089 sbi->ll_foreign_symlink_prefix_size =
1093 sbi->ll_foreign_symlink_prefix_size =
1094 args->to - args->from + 1;
1095 OBD_ALLOC_POST(sbi->ll_foreign_symlink_prefix,
1096 sbi->ll_foreign_symlink_prefix_size,
1099 OBD_FREE(old, old_len);
1101 /* enable foreign symlink support */
1102 set_bit(token, sbi->ll_flags);
1104 LCONSOLE_ERROR_MSG(0x152,
1105 "invalid %s option\n", s1);
1116 void ll_lli_init(struct ll_inode_info *lli)
1118 lli->lli_inode_magic = LLI_INODE_MAGIC;
1120 rwlock_init(&lli->lli_lock);
1121 lli->lli_posix_acl = NULL;
1122 /* Do not set lli_fid, it has been initialized already. */
1123 fid_zero(&lli->lli_pfid);
1124 lli->lli_mds_read_och = NULL;
1125 lli->lli_mds_write_och = NULL;
1126 lli->lli_mds_exec_och = NULL;
1127 lli->lli_open_fd_read_count = 0;
1128 lli->lli_open_fd_write_count = 0;
1129 lli->lli_open_fd_exec_count = 0;
1130 mutex_init(&lli->lli_och_mutex);
1131 spin_lock_init(&lli->lli_agl_lock);
1132 spin_lock_init(&lli->lli_layout_lock);
1133 ll_layout_version_set(lli, CL_LAYOUT_GEN_NONE);
1134 lli->lli_clob = NULL;
1136 init_rwsem(&lli->lli_xattrs_list_rwsem);
1137 mutex_init(&lli->lli_xattrs_enq_lock);
1139 LASSERT(lli->lli_vfs_inode.i_mode != 0);
1140 if (S_ISDIR(lli->lli_vfs_inode.i_mode)) {
1141 lli->lli_opendir_key = NULL;
1142 lli->lli_sai = NULL;
1143 spin_lock_init(&lli->lli_sa_lock);
1144 lli->lli_opendir_pid = 0;
1145 lli->lli_sa_enabled = 0;
1146 init_rwsem(&lli->lli_lsm_sem);
1148 mutex_init(&lli->lli_size_mutex);
1149 mutex_init(&lli->lli_setattr_mutex);
1150 lli->lli_symlink_name = NULL;
1151 ll_trunc_sem_init(&lli->lli_trunc_sem);
1152 range_lock_tree_init(&lli->lli_write_tree);
1153 init_rwsem(&lli->lli_glimpse_sem);
1154 lli->lli_glimpse_time = ktime_set(0, 0);
1155 INIT_LIST_HEAD(&lli->lli_agl_list);
1156 lli->lli_agl_index = 0;
1157 lli->lli_async_rc = 0;
1158 spin_lock_init(&lli->lli_heat_lock);
1159 obd_heat_clear(lli->lli_heat_instances, OBD_HEAT_COUNT);
1160 lli->lli_heat_flags = 0;
1161 mutex_init(&lli->lli_pcc_lock);
1162 lli->lli_pcc_state = PCC_STATE_FL_NONE;
1163 lli->lli_pcc_inode = NULL;
1164 lli->lli_pcc_dsflags = PCC_DATASET_INVALID;
1165 lli->lli_pcc_generation = 0;
1166 mutex_init(&lli->lli_group_mutex);
1167 lli->lli_group_users = 0;
1168 lli->lli_group_gid = 0;
1170 mutex_init(&lli->lli_layout_mutex);
1171 memset(lli->lli_jobid, 0, sizeof(lli->lli_jobid));
1172 /* ll_cl_context initialize */
1173 INIT_LIST_HEAD(&lli->lli_lccs);
1176 #define MAX_STRING_SIZE 128
1178 #ifndef HAVE_SUPER_SETUP_BDI_NAME
1180 #define LSI_BDI_INITIALIZED 0x00400000
1182 #ifndef HAVE_BDI_CAP_MAP_COPY
1183 # define BDI_CAP_MAP_COPY 0
1186 static int super_setup_bdi_name(struct super_block *sb, char *fmt, ...)
1188 struct lustre_sb_info *lsi = s2lsi(sb);
1189 char buf[MAX_STRING_SIZE];
1193 err = bdi_init(&lsi->lsi_bdi);
1197 lsi->lsi_flags |= LSI_BDI_INITIALIZED;
1198 lsi->lsi_bdi.capabilities = BDI_CAP_MAP_COPY;
1199 lsi->lsi_bdi.name = "lustre";
1200 va_start(args, fmt);
1201 vsnprintf(buf, MAX_STRING_SIZE, fmt, args);
1203 err = bdi_register(&lsi->lsi_bdi, NULL, "%s", buf);
1206 sb->s_bdi = &lsi->lsi_bdi;
1210 #endif /* !HAVE_SUPER_SETUP_BDI_NAME */
1212 int ll_fill_super(struct super_block *sb)
1214 struct lustre_profile *lprof = NULL;
1215 struct lustre_sb_info *lsi = s2lsi(sb);
1216 struct ll_sb_info *sbi = NULL;
1217 char *dt = NULL, *md = NULL;
1218 char *profilenm = get_profile_name(sb);
1219 struct config_llog_instance *cfg;
1220 /* %p for void* in printf needs 16+2 characters: 0xffffffffffffffff */
1221 const int instlen = LUSTRE_MAXINSTANCE + 2;
1222 unsigned long cfg_instance = ll_get_cfg_instance(sb);
1223 char name[MAX_STRING_SIZE];
1232 /* for ASLR, to map between cfg_instance and hashed ptr */
1233 CDEBUG(D_VFSTRACE, "VFS Op: cfg_instance %s-%016lx (sb %p)\n",
1234 profilenm, cfg_instance, sb);
1236 OBD_RACE(OBD_FAIL_LLITE_RACE_MOUNT);
1240 GOTO(out_free_cfg, err = -ENOMEM);
1242 /* client additional sb info */
1243 lsi->lsi_llsbi = sbi = ll_init_sbi();
1245 GOTO(out_free_cfg, err = PTR_ERR(sbi));
1247 err = ll_options(lsi->lsi_lmd->lmd_opts, sb);
1249 GOTO(out_free_cfg, err);
1251 /* kernel >= 2.6.38 store dentry operations in sb->s_d_op. */
1252 sb->s_d_op = &ll_d_ops;
1255 generate_random_uuid(uuid.b);
1256 snprintf(sbi->ll_sb_uuid.uuid, sizeof(sbi->ll_sb_uuid), "%pU", uuid.b);
1258 CDEBUG(D_CONFIG, "llite sb uuid: %s\n", sbi->ll_sb_uuid.uuid);
1261 len = strlen(profilenm);
1262 ptr = strrchr(profilenm, '-');
1263 if (ptr && (strcmp(ptr, "-client") == 0))
1266 if (len > LUSTRE_MAXFSNAME) {
1267 if (unlikely(len >= MAX_STRING_SIZE))
1268 len = MAX_STRING_SIZE - 1;
1269 strncpy(name, profilenm, len);
1271 err = -ENAMETOOLONG;
1272 CERROR("%s: fsname longer than %u characters: rc = %d\n",
1273 name, LUSTRE_MAXFSNAME, err);
1274 GOTO(out_free_cfg, err);
1276 strncpy(sbi->ll_fsname, profilenm, len);
1277 sbi->ll_fsname[len] = '\0';
1280 snprintf(name, sizeof(name), "%.*s-%016lx", len,
1281 profilenm, cfg_instance);
1283 err = super_setup_bdi_name(sb, "%s", name);
1285 GOTO(out_free_cfg, err);
1287 /* Call ll_debugfs_register_super() before lustre_process_log()
1288 * so that "llite.*.*" params can be processed correctly.
1290 err = ll_debugfs_register_super(sb, name);
1292 CERROR("%s: could not register mountpoint in llite: rc = %d\n",
1293 sbi->ll_fsname, err);
1297 /* The cfg_instance is a value unique to this super, in case some
1298 * joker tries to mount the same fs at two mount points.
1300 cfg->cfg_instance = cfg_instance;
1301 cfg->cfg_uuid = lsi->lsi_llsbi->ll_sb_uuid;
1302 cfg->cfg_callback = class_config_llog_handler;
1303 cfg->cfg_sub_clds = CONFIG_SUB_CLIENT;
1304 /* set up client obds */
1305 err = lustre_process_log(sb, profilenm, cfg);
1307 GOTO(out_debugfs, err);
1309 /* Profile set with LCFG_MOUNTOPT so we can find our mdc and osc obds */
1310 lprof = class_get_profile(profilenm);
1311 if (lprof == NULL) {
1312 LCONSOLE_ERROR_MSG(0x156, "The client profile '%s' could not be"
1313 " read from the MGS. Does that filesystem "
1314 "exist?\n", profilenm);
1315 GOTO(out_debugfs, err = -EINVAL);
1317 CDEBUG(D_CONFIG, "Found profile %s: mdc=%s osc=%s\n", profilenm,
1318 lprof->lp_md, lprof->lp_dt);
1320 dt_len = strlen(lprof->lp_dt) + instlen + 2;
1321 OBD_ALLOC(dt, dt_len);
1323 GOTO(out_profile, err = -ENOMEM);
1324 snprintf(dt, dt_len - 1, "%s-%016lx", lprof->lp_dt, cfg_instance);
1326 md_len = strlen(lprof->lp_md) + instlen + 2;
1327 OBD_ALLOC(md, md_len);
1329 GOTO(out_free_dt, err = -ENOMEM);
1330 snprintf(md, md_len - 1, "%s-%016lx", lprof->lp_md, cfg_instance);
1332 /* connections, registrations, sb setup */
1333 err = client_common_fill_super(sb, md, dt);
1335 GOTO(out_free_md, err);
1337 sbi->ll_client_common_fill_super_succeeded = 1;
1341 OBD_FREE(md, md_len);
1344 OBD_FREE(dt, dt_len);
1347 class_put_profile(lprof);
1350 ll_debugfs_unregister_super(sb);
1357 else if (test_bit(LL_SBI_VERBOSE, sbi->ll_flags))
1358 LCONSOLE_WARN("Mounted %s\n", profilenm);
1360 } /* ll_fill_super */
1362 void ll_put_super(struct super_block *sb)
1364 struct config_llog_instance cfg, params_cfg;
1365 struct obd_device *obd;
1366 struct lustre_sb_info *lsi = s2lsi(sb);
1367 struct ll_sb_info *sbi = ll_s2sbi(sb);
1368 char *profilenm = get_profile_name(sb);
1369 unsigned long cfg_instance = ll_get_cfg_instance(sb);
1371 int next, force = 1, rc = 0;
1375 GOTO(out_no_sbi, 0);
1377 /* Should replace instance_id with something better for ASLR */
1378 CDEBUG(D_VFSTRACE, "VFS Op: cfg_instance %s-%016lx (sb %p)\n",
1379 profilenm, cfg_instance, sb);
1381 cfg.cfg_instance = cfg_instance;
1382 lustre_end_log(sb, profilenm, &cfg);
1384 params_cfg.cfg_instance = cfg_instance;
1385 lustre_end_log(sb, PARAMS_FILENAME, ¶ms_cfg);
1387 if (sbi->ll_md_exp) {
1388 obd = class_exp2obd(sbi->ll_md_exp);
1390 force = obd->obd_force;
1393 /* Wait for unstable pages to be committed to stable storage */
1395 rc = l_wait_event_abortable(
1396 sbi->ll_cache->ccc_unstable_waitq,
1397 atomic_long_read(&sbi->ll_cache->ccc_unstable_nr) == 0);
1400 ccc_count = atomic_long_read(&sbi->ll_cache->ccc_unstable_nr);
1401 if (force == 0 && rc != -ERESTARTSYS)
1402 LASSERTF(ccc_count == 0, "count: %li\n", ccc_count);
1404 /* We need to set force before the lov_disconnect in
1405 * lustre_common_put_super, since l_d cleans up osc's as well.
1409 while ((obd = class_devices_in_group(&sbi->ll_sb_uuid,
1411 obd->obd_force = force;
1415 if (sbi->ll_client_common_fill_super_succeeded) {
1416 /* Only if client_common_fill_super succeeded */
1417 client_common_put_super(sb);
1421 while ((obd = class_devices_in_group(&sbi->ll_sb_uuid, &next)))
1422 class_manual_cleanup(obd);
1424 if (test_bit(LL_SBI_VERBOSE, sbi->ll_flags))
1425 LCONSOLE_WARN("Unmounted %s\n", profilenm ? profilenm : "");
1428 class_del_profile(profilenm);
1430 #ifndef HAVE_SUPER_SETUP_BDI_NAME
1431 if (lsi->lsi_flags & LSI_BDI_INITIALIZED) {
1432 bdi_destroy(&lsi->lsi_bdi);
1433 lsi->lsi_flags &= ~LSI_BDI_INITIALIZED;
1438 lsi->lsi_llsbi = NULL;
1440 lustre_common_put_super(sb);
1442 cl_env_cache_purge(~0);
1445 } /* client_put_super */
1447 struct inode *ll_inode_from_resource_lock(struct ldlm_lock *lock)
1449 struct inode *inode = NULL;
1451 /* NOTE: we depend on atomic igrab() -bzzz */
1452 lock_res_and_lock(lock);
1453 if (lock->l_resource->lr_lvb_inode) {
1454 struct ll_inode_info * lli;
1455 lli = ll_i2info(lock->l_resource->lr_lvb_inode);
1456 if (lli->lli_inode_magic == LLI_INODE_MAGIC) {
1457 inode = igrab(lock->l_resource->lr_lvb_inode);
1459 inode = lock->l_resource->lr_lvb_inode;
1460 LDLM_DEBUG_LIMIT(inode->i_state & I_FREEING ? D_INFO :
1461 D_WARNING, lock, "lr_lvb_inode %p is "
1462 "bogus: magic %08x",
1463 lock->l_resource->lr_lvb_inode,
1464 lli->lli_inode_magic);
1468 unlock_res_and_lock(lock);
1472 void ll_dir_clear_lsm_md(struct inode *inode)
1474 struct ll_inode_info *lli = ll_i2info(inode);
1476 LASSERT(S_ISDIR(inode->i_mode));
1478 if (lli->lli_lsm_md) {
1479 lmv_free_memmd(lli->lli_lsm_md);
1480 lli->lli_lsm_md = NULL;
1483 if (lli->lli_default_lsm_md) {
1484 lmv_free_memmd(lli->lli_default_lsm_md);
1485 lli->lli_default_lsm_md = NULL;
1489 static struct inode *ll_iget_anon_dir(struct super_block *sb,
1490 const struct lu_fid *fid,
1491 struct lustre_md *md)
1493 struct ll_sb_info *sbi = ll_s2sbi(sb);
1494 struct ll_inode_info *lli;
1495 struct mdt_body *body = md->body;
1496 struct inode *inode;
1502 ino = cl_fid_build_ino(fid, test_bit(LL_SBI_32BIT_API, sbi->ll_flags));
1503 inode = iget_locked(sb, ino);
1504 if (inode == NULL) {
1505 CERROR("%s: failed get simple inode "DFID": rc = -ENOENT\n",
1506 sbi->ll_fsname, PFID(fid));
1507 RETURN(ERR_PTR(-ENOENT));
1510 lli = ll_i2info(inode);
1511 if (inode->i_state & I_NEW) {
1512 inode->i_mode = (inode->i_mode & ~S_IFMT) |
1513 (body->mbo_mode & S_IFMT);
1514 LASSERTF(S_ISDIR(inode->i_mode), "Not slave inode "DFID"\n",
1517 inode->i_mtime.tv_sec = 0;
1518 inode->i_atime.tv_sec = 0;
1519 inode->i_ctime.tv_sec = 0;
1522 #ifdef HAVE_BACKING_DEV_INFO
1523 /* initializing backing dev info. */
1524 inode->i_mapping->backing_dev_info =
1525 &s2lsi(inode->i_sb)->lsi_bdi;
1527 inode->i_op = &ll_dir_inode_operations;
1528 inode->i_fop = &ll_dir_operations;
1529 lli->lli_fid = *fid;
1532 /* master object FID */
1533 lli->lli_pfid = body->mbo_fid1;
1534 CDEBUG(D_INODE, "lli %p slave "DFID" master "DFID"\n",
1535 lli, PFID(fid), PFID(&lli->lli_pfid));
1536 unlock_new_inode(inode);
1538 /* in directory restripe/auto-split, a directory will be
1539 * transformed to a stripe if it's plain, set its pfid here,
1540 * otherwise ll_lock_cancel_bits() can't find the master inode.
1542 lli->lli_pfid = body->mbo_fid1;
1548 static int ll_init_lsm_md(struct inode *inode, struct lustre_md *md)
1551 struct lmv_stripe_md *lsm = md->lmv;
1552 struct ll_inode_info *lli = ll_i2info(inode);
1555 LASSERT(lsm != NULL);
1557 CDEBUG(D_INODE, "%s: "DFID" set dir layout:\n",
1558 ll_i2sbi(inode)->ll_fsname, PFID(&lli->lli_fid));
1559 lsm_md_dump(D_INODE, lsm);
1561 if (!lmv_dir_striped(lsm))
1564 /* XXX sigh, this lsm_root initialization should be in
1565 * LMV layer, but it needs ll_iget right now, so we
1566 * put this here right now. */
1567 for (i = 0; i < lsm->lsm_md_stripe_count; i++) {
1568 fid = &lsm->lsm_md_oinfo[i].lmo_fid;
1569 LASSERT(lsm->lsm_md_oinfo[i].lmo_root == NULL);
1571 if (!fid_is_sane(fid))
1574 /* Unfortunately ll_iget will call ll_update_inode,
1575 * where the initialization of slave inode is slightly
1576 * different, so it reset lsm_md to NULL to avoid
1577 * initializing lsm for slave inode. */
1578 lsm->lsm_md_oinfo[i].lmo_root =
1579 ll_iget_anon_dir(inode->i_sb, fid, md);
1580 if (IS_ERR(lsm->lsm_md_oinfo[i].lmo_root)) {
1581 int rc = PTR_ERR(lsm->lsm_md_oinfo[i].lmo_root);
1583 lsm->lsm_md_oinfo[i].lmo_root = NULL;
1585 iput(lsm->lsm_md_oinfo[i].lmo_root);
1586 lsm->lsm_md_oinfo[i].lmo_root = NULL;
1592 lli->lli_lsm_md = lsm;
1597 static void ll_update_default_lsm_md(struct inode *inode, struct lustre_md *md)
1599 struct ll_inode_info *lli = ll_i2info(inode);
1601 if (!md->default_lmv) {
1602 /* clear default lsm */
1603 if (lli->lli_default_lsm_md) {
1604 down_write(&lli->lli_lsm_sem);
1605 if (lli->lli_default_lsm_md) {
1606 lmv_free_memmd(lli->lli_default_lsm_md);
1607 lli->lli_default_lsm_md = NULL;
1609 up_write(&lli->lli_lsm_sem);
1614 if (lli->lli_default_lsm_md) {
1615 /* do nonthing if default lsm isn't changed */
1616 down_read(&lli->lli_lsm_sem);
1617 if (lli->lli_default_lsm_md &&
1618 lsm_md_eq(lli->lli_default_lsm_md, md->default_lmv)) {
1619 up_read(&lli->lli_lsm_sem);
1622 up_read(&lli->lli_lsm_sem);
1625 down_write(&lli->lli_lsm_sem);
1626 if (lli->lli_default_lsm_md)
1627 lmv_free_memmd(lli->lli_default_lsm_md);
1628 lli->lli_default_lsm_md = md->default_lmv;
1629 lsm_md_dump(D_INODE, md->default_lmv);
1630 md->default_lmv = NULL;
1631 up_write(&lli->lli_lsm_sem);
1634 static int ll_update_lsm_md(struct inode *inode, struct lustre_md *md)
1636 struct ll_inode_info *lli = ll_i2info(inode);
1637 struct lmv_stripe_md *lsm = md->lmv;
1638 struct cl_attr *attr;
1643 LASSERT(S_ISDIR(inode->i_mode));
1644 CDEBUG(D_INODE, "update lsm %p of "DFID"\n", lli->lli_lsm_md,
1645 PFID(ll_inode2fid(inode)));
1647 /* update default LMV */
1648 if (md->default_lmv)
1649 ll_update_default_lsm_md(inode, md);
1651 /* after dir migration/restripe, a stripe may be turned into a
1652 * directory, in this case, zero out its lli_pfid.
1654 if (unlikely(fid_is_norm(&lli->lli_pfid)))
1655 fid_zero(&lli->lli_pfid);
1658 * no striped information from request, lustre_md from req does not
1659 * include stripeEA, see ll_md_setattr()
1665 * normally dir layout doesn't change, only take read lock to check
1666 * that to avoid blocking other MD operations.
1668 down_read(&lli->lli_lsm_sem);
1670 /* some current lookup initialized lsm, and unchanged */
1671 if (lli->lli_lsm_md && lsm_md_eq(lli->lli_lsm_md, lsm))
1672 GOTO(unlock, rc = 0);
1674 /* if dir layout doesn't match, check whether version is increased,
1675 * which means layout is changed, this happens in dir split/merge and
1678 * foreign LMV should not change.
1680 if (lli->lli_lsm_md && lmv_dir_striped(lli->lli_lsm_md) &&
1681 lsm->lsm_md_layout_version <=
1682 lli->lli_lsm_md->lsm_md_layout_version) {
1683 CERROR("%s: "DFID" dir layout mismatch:\n",
1684 ll_i2sbi(inode)->ll_fsname, PFID(&lli->lli_fid));
1685 lsm_md_dump(D_ERROR, lli->lli_lsm_md);
1686 lsm_md_dump(D_ERROR, lsm);
1687 GOTO(unlock, rc = -EINVAL);
1690 up_read(&lli->lli_lsm_sem);
1691 down_write(&lli->lli_lsm_sem);
1692 /* clear existing lsm */
1693 if (lli->lli_lsm_md) {
1694 lmv_free_memmd(lli->lli_lsm_md);
1695 lli->lli_lsm_md = NULL;
1698 rc = ll_init_lsm_md(inode, md);
1699 up_write(&lli->lli_lsm_sem);
1704 /* set md->lmv to NULL, so the following free lustre_md will not free
1709 /* md_merge_attr() may take long, since lsm is already set, switch to
1712 down_read(&lli->lli_lsm_sem);
1714 if (!lmv_dir_striped(lli->lli_lsm_md))
1715 GOTO(unlock, rc = 0);
1717 OBD_ALLOC_PTR(attr);
1719 GOTO(unlock, rc = -ENOMEM);
1721 /* validate the lsm */
1722 rc = md_merge_attr(ll_i2mdexp(inode), lli->lli_lsm_md, attr,
1723 ll_md_blocking_ast);
1725 if (md->body->mbo_valid & OBD_MD_FLNLINK)
1726 md->body->mbo_nlink = attr->cat_nlink;
1727 if (md->body->mbo_valid & OBD_MD_FLSIZE)
1728 md->body->mbo_size = attr->cat_size;
1729 if (md->body->mbo_valid & OBD_MD_FLATIME)
1730 md->body->mbo_atime = attr->cat_atime;
1731 if (md->body->mbo_valid & OBD_MD_FLCTIME)
1732 md->body->mbo_ctime = attr->cat_ctime;
1733 if (md->body->mbo_valid & OBD_MD_FLMTIME)
1734 md->body->mbo_mtime = attr->cat_mtime;
1740 up_read(&lli->lli_lsm_sem);
1745 void ll_clear_inode(struct inode *inode)
1747 struct ll_inode_info *lli = ll_i2info(inode);
1748 struct ll_sb_info *sbi = ll_i2sbi(inode);
1752 CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p)\n",
1753 PFID(ll_inode2fid(inode)), inode);
1755 if (S_ISDIR(inode->i_mode)) {
1756 /* these should have been cleared in ll_file_release */
1757 LASSERT(lli->lli_opendir_key == NULL);
1758 LASSERT(lli->lli_sai == NULL);
1759 LASSERT(lli->lli_opendir_pid == 0);
1761 pcc_inode_free(inode);
1764 md_null_inode(sbi->ll_md_exp, ll_inode2fid(inode));
1766 LASSERT(!lli->lli_open_fd_write_count);
1767 LASSERT(!lli->lli_open_fd_read_count);
1768 LASSERT(!lli->lli_open_fd_exec_count);
1770 if (lli->lli_mds_write_och)
1771 ll_md_real_close(inode, FMODE_WRITE);
1772 if (lli->lli_mds_exec_och)
1773 ll_md_real_close(inode, FMODE_EXEC);
1774 if (lli->lli_mds_read_och)
1775 ll_md_real_close(inode, FMODE_READ);
1777 if (S_ISLNK(inode->i_mode) && lli->lli_symlink_name) {
1778 OBD_FREE(lli->lli_symlink_name,
1779 strlen(lli->lli_symlink_name) + 1);
1780 lli->lli_symlink_name = NULL;
1783 ll_xattr_cache_destroy(inode);
1785 forget_all_cached_acls(inode);
1787 lli->lli_inode_magic = LLI_INODE_DEAD;
1789 if (S_ISDIR(inode->i_mode))
1790 ll_dir_clear_lsm_md(inode);
1791 else if (S_ISREG(inode->i_mode) && !is_bad_inode(inode))
1792 LASSERT(list_empty(&lli->lli_agl_list));
1795 * XXX This has to be done before lsm is freed below, because
1796 * cl_object still uses inode lsm.
1798 cl_inode_fini(inode);
1800 llcrypt_put_encryption_info(inode);
1805 static int ll_md_setattr(struct dentry *dentry, struct md_op_data *op_data)
1807 struct lustre_md md;
1808 struct inode *inode = dentry->d_inode;
1809 struct ll_sb_info *sbi = ll_i2sbi(inode);
1810 struct ptlrpc_request *request = NULL;
1815 op_data = ll_prep_md_op_data(op_data, inode, NULL, NULL, 0, 0,
1816 LUSTRE_OPC_ANY, NULL);
1817 if (IS_ERR(op_data))
1818 RETURN(PTR_ERR(op_data));
1820 /* If this is a chgrp of a regular file, we want to reserve enough
1821 * quota to cover the entire file size.
1823 if (S_ISREG(inode->i_mode) && op_data->op_attr.ia_valid & ATTR_GID &&
1824 from_kgid(&init_user_ns, op_data->op_attr.ia_gid) !=
1825 from_kgid(&init_user_ns, inode->i_gid)) {
1826 op_data->op_xvalid |= OP_XVALID_BLOCKS;
1827 op_data->op_attr_blocks = inode->i_blocks;
1831 rc = md_setattr(sbi->ll_md_exp, op_data, NULL, 0, &request);
1833 ptlrpc_req_finished(request);
1834 if (rc == -ENOENT) {
1836 /* Unlinked special device node? Or just a race?
1837 * Pretend we done everything. */
1838 if (!S_ISREG(inode->i_mode) &&
1839 !S_ISDIR(inode->i_mode)) {
1840 ia_valid = op_data->op_attr.ia_valid;
1841 op_data->op_attr.ia_valid &= ~TIMES_SET_FLAGS;
1842 rc = simple_setattr(dentry, &op_data->op_attr);
1843 op_data->op_attr.ia_valid = ia_valid;
1845 } else if (rc != -EPERM && rc != -EACCES && rc != -ETXTBSY) {
1846 CERROR("md_setattr fails: rc = %d\n", rc);
1851 rc = md_get_lustre_md(sbi->ll_md_exp, &request->rq_pill, sbi->ll_dt_exp,
1852 sbi->ll_md_exp, &md);
1854 ptlrpc_req_finished(request);
1858 ia_valid = op_data->op_attr.ia_valid;
1859 /* inode size will be in ll_setattr_ost, can't do it now since dirty
1860 * cache is not cleared yet. */
1861 op_data->op_attr.ia_valid &= ~(TIMES_SET_FLAGS | ATTR_SIZE);
1862 if (S_ISREG(inode->i_mode))
1864 rc = simple_setattr(dentry, &op_data->op_attr);
1865 if (S_ISREG(inode->i_mode))
1866 inode_unlock(inode);
1867 op_data->op_attr.ia_valid = ia_valid;
1869 rc = ll_update_inode(inode, &md);
1870 ptlrpc_req_finished(request);
1876 * Zero portion of page that is part of @inode.
1877 * This implies, if necessary:
1878 * - taking cl_lock on range corresponding to concerned page
1879 * - grabbing vm page
1880 * - associating cl_page
1881 * - proceeding to clio read
1882 * - zeroing range in page
1883 * - proceeding to cl_page flush
1884 * - releasing cl_lock
1886 * \param[in] inode inode
1887 * \param[in] index page index
1888 * \param[in] offset offset in page to start zero from
1889 * \param[in] len len to zero
1891 * \retval 0 on success
1892 * \retval negative errno on failure
1894 int ll_io_zero_page(struct inode *inode, pgoff_t index, pgoff_t offset,
1897 struct ll_inode_info *lli = ll_i2info(inode);
1898 struct cl_object *clob = lli->lli_clob;
1900 struct lu_env *env = NULL;
1901 struct cl_io *io = NULL;
1902 struct cl_page *clpage = NULL;
1903 struct page *vmpage = NULL;
1904 unsigned from = index << PAGE_SHIFT;
1905 struct cl_lock *lock = NULL;
1906 struct cl_lock_descr *descr = NULL;
1907 struct cl_2queue *queue = NULL;
1908 struct cl_sync_io *anchor = NULL;
1909 bool holdinglock = false;
1910 bool lockedbymyself = true;
1915 env = cl_env_get(&refcheck);
1917 RETURN(PTR_ERR(env));
1919 io = vvp_env_thread_io(env);
1921 rc = cl_io_rw_init(env, io, CIT_WRITE, from, PAGE_SIZE);
1925 lock = vvp_env_lock(env);
1926 descr = &lock->cll_descr;
1927 descr->cld_obj = io->ci_obj;
1928 descr->cld_start = cl_index(io->ci_obj, from);
1929 descr->cld_end = cl_index(io->ci_obj, from + PAGE_SIZE - 1);
1930 descr->cld_mode = CLM_WRITE;
1931 descr->cld_enq_flags = CEF_MUST | CEF_NONBLOCK;
1933 /* request lock for page */
1934 rc = cl_lock_request(env, io, lock);
1935 /* -ECANCELED indicates a matching lock with a different extent
1936 * was already present, and -EEXIST indicates a matching lock
1937 * on exactly the same extent was already present.
1938 * In both cases it means we are covered.
1940 if (rc == -ECANCELED || rc == -EEXIST)
1948 vmpage = grab_cache_page_nowait(inode->i_mapping, index);
1950 GOTO(rellock, rc = -EOPNOTSUPP);
1952 if (!PageDirty(vmpage)) {
1953 /* associate cl_page */
1954 clpage = cl_page_find(env, clob, vmpage->index,
1955 vmpage, CPT_CACHEABLE);
1957 GOTO(pagefini, rc = PTR_ERR(clpage));
1959 cl_page_assume(env, io, clpage);
1962 if (!PageUptodate(vmpage) && !PageDirty(vmpage) &&
1963 !PageWriteback(vmpage)) {
1965 /* set PagePrivate2 to detect special case of empty page
1966 * in osc_brw_fini_request()
1968 SetPagePrivate2(vmpage);
1969 rc = ll_io_read_page(env, io, clpage, NULL);
1970 if (!PagePrivate2(vmpage))
1971 /* PagePrivate2 was cleared in osc_brw_fini_request()
1972 * meaning we read an empty page. In this case, in order
1973 * to avoid allocating unnecessary block in truncated
1974 * file, we must not zero and write as below. Subsequent
1975 * server-side truncate will handle things correctly.
1977 GOTO(clpfini, rc = 0);
1978 ClearPagePrivate2(vmpage);
1981 lockedbymyself = trylock_page(vmpage);
1982 cl_page_assume(env, io, clpage);
1985 /* zero range in page */
1986 zero_user(vmpage, offset, len);
1988 if (holdinglock && clpage) {
1989 /* explicitly write newly modified page */
1990 queue = &io->ci_queue;
1991 cl_2queue_init(queue);
1992 anchor = &vvp_env_info(env)->vti_anchor;
1993 cl_sync_io_init(anchor, 1);
1994 clpage->cp_sync_io = anchor;
1995 cl_2queue_add(queue, clpage, true);
1996 rc = cl_io_submit_rw(env, io, CRT_WRITE, queue);
1998 GOTO(queuefini1, rc);
1999 rc = cl_sync_io_wait(env, anchor, 0);
2001 GOTO(queuefini2, rc);
2002 cl_page_assume(env, io, clpage);
2005 cl_2queue_discard(env, io, queue);
2007 cl_2queue_disown(env, io, queue);
2008 cl_2queue_fini(env, queue);
2013 cl_page_put(env, clpage);
2015 if (lockedbymyself) {
2016 unlock_page(vmpage);
2021 cl_lock_release(env, lock);
2023 cl_io_fini(env, io);
2026 cl_env_put(env, &refcheck);
2031 /* If this inode has objects allocated to it (lsm != NULL), then the OST
2032 * object(s) determine the file size and mtime. Otherwise, the MDS will
2033 * keep these values until such a time that objects are allocated for it.
2034 * We do the MDS operations first, as it is checking permissions for us.
2035 * We don't to the MDS RPC if there is nothing that we want to store there,
2036 * otherwise there is no harm in updating mtime/atime on the MDS if we are
2037 * going to do an RPC anyways.
2039 * If we are doing a truncate, we will send the mtime and ctime updates
2040 * to the OST with the punch RPC, otherwise we do an explicit setattr RPC.
2041 * I don't believe it is possible to get e.g. ATTR_MTIME_SET and ATTR_SIZE
2044 * In case of HSMimport, we only set attr on MDS.
2046 int ll_setattr_raw(struct dentry *dentry, struct iattr *attr,
2047 enum op_xvalid xvalid, bool hsm_import)
2049 struct inode *inode = dentry->d_inode;
2050 struct ll_inode_info *lli = ll_i2info(inode);
2051 struct md_op_data *op_data = NULL;
2052 ktime_t kstart = ktime_get();
2057 CDEBUG(D_VFSTRACE, "%s: setattr inode "DFID"(%p) from %llu to %llu, "
2058 "valid %x, hsm_import %d\n",
2059 ll_i2sbi(inode)->ll_fsname, PFID(&lli->lli_fid),
2060 inode, i_size_read(inode), attr->ia_size, attr->ia_valid,
2063 if (attr->ia_valid & ATTR_SIZE) {
2064 /* Check new size against VFS/VM file size limit and rlimit */
2065 rc = inode_newsize_ok(inode, attr->ia_size);
2069 /* The maximum Lustre file size is variable, based on the
2070 * OST maximum object size and number of stripes. This
2071 * needs another check in addition to the VFS check above. */
2072 if (attr->ia_size > ll_file_maxbytes(inode)) {
2073 CDEBUG(D_INODE,"file "DFID" too large %llu > %llu\n",
2074 PFID(&lli->lli_fid), attr->ia_size,
2075 ll_file_maxbytes(inode));
2079 attr->ia_valid |= ATTR_MTIME | ATTR_CTIME;
2082 /* POSIX: check before ATTR_*TIME_SET set (from inode_change_ok) */
2083 if (attr->ia_valid & TIMES_SET_FLAGS) {
2084 if ((!uid_eq(current_fsuid(), inode->i_uid)) &&
2085 !capable(CAP_FOWNER))
2089 /* We mark all of the fields "set" so MDS/OST does not re-set them */
2090 if (!(xvalid & OP_XVALID_CTIME_SET) &&
2091 (attr->ia_valid & ATTR_CTIME)) {
2092 attr->ia_ctime = current_time(inode);
2093 xvalid |= OP_XVALID_CTIME_SET;
2095 if (!(attr->ia_valid & ATTR_ATIME_SET) &&
2096 (attr->ia_valid & ATTR_ATIME)) {
2097 attr->ia_atime = current_time(inode);
2098 attr->ia_valid |= ATTR_ATIME_SET;
2100 if (!(attr->ia_valid & ATTR_MTIME_SET) &&
2101 (attr->ia_valid & ATTR_MTIME)) {
2102 attr->ia_mtime = current_time(inode);
2103 attr->ia_valid |= ATTR_MTIME_SET;
2106 if (attr->ia_valid & (ATTR_MTIME | ATTR_CTIME))
2107 CDEBUG(D_INODE, "setting mtime %lld, ctime %lld, now = %lld\n",
2108 (s64)attr->ia_mtime.tv_sec, (s64)attr->ia_ctime.tv_sec,
2109 ktime_get_real_seconds());
2111 if (S_ISREG(inode->i_mode))
2112 inode_unlock(inode);
2114 /* We always do an MDS RPC, even if we're only changing the size;
2115 * only the MDS knows whether truncate() should fail with -ETXTBUSY */
2117 OBD_ALLOC_PTR(op_data);
2118 if (op_data == NULL)
2119 GOTO(out, rc = -ENOMEM);
2121 if (!hsm_import && attr->ia_valid & ATTR_SIZE) {
2122 /* If we are changing file size, file content is
2123 * modified, flag it.
2125 xvalid |= OP_XVALID_OWNEROVERRIDE;
2126 op_data->op_bias |= MDS_DATA_MODIFIED;
2127 clear_bit(LLIF_DATA_MODIFIED, &lli->lli_flags);
2130 if (attr->ia_valid & ATTR_FILE) {
2131 struct ll_file_data *fd = attr->ia_file->private_data;
2133 if (fd->fd_lease_och)
2134 op_data->op_bias |= MDS_TRUNC_KEEP_LEASE;
2137 op_data->op_attr = *attr;
2138 op_data->op_xvalid = xvalid;
2140 rc = ll_md_setattr(dentry, op_data);
2144 if (!S_ISREG(inode->i_mode) || hsm_import)
2147 if (attr->ia_valid & (ATTR_SIZE | ATTR_ATIME | ATTR_ATIME_SET |
2148 ATTR_MTIME | ATTR_MTIME_SET | ATTR_CTIME) ||
2149 xvalid & OP_XVALID_CTIME_SET) {
2150 bool cached = false;
2152 rc = pcc_inode_setattr(inode, attr, &cached);
2155 CERROR("%s: PCC inode "DFID" setattr failed: "
2157 ll_i2sbi(inode)->ll_fsname,
2158 PFID(&lli->lli_fid), rc);
2162 unsigned int flags = 0;
2164 /* For truncate and utimes sending attributes to OSTs,
2165 * setting mtime/atime to the past will be performed
2166 * under PW [0:EOF] extent lock (new_size:EOF for
2167 * truncate). It may seem excessive to send mtime/atime
2168 * updates to OSTs when not setting times to past, but
2169 * it is necessary due to possible time
2170 * de-synchronization between MDT inode and OST objects
2172 if (S_ISREG(inode->i_mode) && IS_ENCRYPTED(inode)) {
2173 xvalid |= OP_XVALID_FLAGS;
2174 flags = LUSTRE_ENCRYPT_FL;
2175 /* Call to ll_io_zero_page is not necessary if
2176 * truncating on PAGE_SIZE boundary, because
2177 * whole pages will be wiped.
2178 * In case of Direct IO, all we need is to set
2181 if (attr->ia_valid & ATTR_SIZE &&
2182 attr->ia_size & ~PAGE_MASK &&
2183 !(attr->ia_valid & ATTR_FILE &&
2184 attr->ia_file->f_flags & O_DIRECT)) {
2186 attr->ia_size & (PAGE_SIZE - 1);
2188 rc = ll_io_zero_page(inode,
2189 attr->ia_size >> PAGE_SHIFT,
2190 offset, PAGE_SIZE - offset);
2195 rc = cl_setattr_ost(lli->lli_clob, attr, xvalid, flags);
2199 /* If the file was restored, it needs to set dirty flag.
2201 * We've already sent MDS_DATA_MODIFIED flag in
2202 * ll_md_setattr() for truncate. However, the MDT refuses to
2203 * set the HS_DIRTY flag on released files, so we have to set
2204 * it again if the file has been restored. Please check how
2205 * LLIF_DATA_MODIFIED is set in vvp_io_setattr_fini().
2207 * Please notice that if the file is not released, the previous
2208 * MDS_DATA_MODIFIED has taken effect and usually
2209 * LLIF_DATA_MODIFIED is not set(see vvp_io_setattr_fini()).
2210 * This way we can save an RPC for common open + trunc
2212 if (test_and_clear_bit(LLIF_DATA_MODIFIED, &lli->lli_flags)) {
2213 struct hsm_state_set hss = {
2214 .hss_valid = HSS_SETMASK,
2215 .hss_setmask = HS_DIRTY,
2219 rc2 = ll_hsm_state_set(inode, &hss);
2220 /* truncate and write can happen at the same time, so that
2221 * the file can be set modified even though the file is not
2222 * restored from released state, and ll_hsm_state_set() is
2223 * not applicable for the file, and rc2 < 0 is normal in this
2226 CDEBUG(D_INFO, DFID "HSM set dirty failed: rc2 = %d\n",
2227 PFID(ll_inode2fid(inode)), rc2);
2232 if (op_data != NULL)
2233 ll_finish_md_op_data(op_data);
2235 if (S_ISREG(inode->i_mode)) {
2237 if ((attr->ia_valid & ATTR_SIZE) && !hsm_import)
2238 inode_dio_wait(inode);
2239 /* Once we've got the i_mutex, it's safe to set the S_NOSEC
2240 * flag. ll_update_inode (called from ll_md_setattr), clears
2241 * inode flags, so there is a gap where S_NOSEC is not set.
2242 * This can cause a writer to take the i_mutex unnecessarily,
2243 * but this is safe to do and should be rare. */
2244 inode_has_no_xattr(inode);
2248 ll_stats_ops_tally(ll_i2sbi(inode), attr->ia_valid & ATTR_SIZE ?
2249 LPROC_LL_TRUNC : LPROC_LL_SETATTR,
2250 ktime_us_delta(ktime_get(), kstart));
2255 int ll_setattr(struct dentry *de, struct iattr *attr)
2257 int mode = de->d_inode->i_mode;
2258 enum op_xvalid xvalid = 0;
2261 rc = llcrypt_prepare_setattr(de, attr);
2265 if ((attr->ia_valid & (ATTR_CTIME|ATTR_SIZE|ATTR_MODE)) ==
2266 (ATTR_CTIME|ATTR_SIZE|ATTR_MODE))
2267 xvalid |= OP_XVALID_OWNEROVERRIDE;
2269 if (((attr->ia_valid & (ATTR_MODE|ATTR_FORCE|ATTR_SIZE)) ==
2270 (ATTR_SIZE|ATTR_MODE)) &&
2271 (((mode & S_ISUID) && !(attr->ia_mode & S_ISUID)) ||
2272 (((mode & (S_ISGID|S_IXGRP)) == (S_ISGID|S_IXGRP)) &&
2273 !(attr->ia_mode & S_ISGID))))
2274 attr->ia_valid |= ATTR_FORCE;
2276 if ((attr->ia_valid & ATTR_MODE) &&
2278 !(attr->ia_mode & S_ISUID) &&
2279 !(attr->ia_valid & ATTR_KILL_SUID))
2280 attr->ia_valid |= ATTR_KILL_SUID;
2282 if ((attr->ia_valid & ATTR_MODE) &&
2283 ((mode & (S_ISGID|S_IXGRP)) == (S_ISGID|S_IXGRP)) &&
2284 !(attr->ia_mode & S_ISGID) &&
2285 !(attr->ia_valid & ATTR_KILL_SGID))
2286 attr->ia_valid |= ATTR_KILL_SGID;
2288 return ll_setattr_raw(de, attr, xvalid, false);
2291 int ll_statfs_internal(struct ll_sb_info *sbi, struct obd_statfs *osfs,
2294 struct obd_statfs obd_osfs = { 0 };
2299 max_age = ktime_get_seconds() - sbi->ll_statfs_max_age;
2301 if (test_bit(LL_SBI_LAZYSTATFS, sbi->ll_flags))
2302 flags |= OBD_STATFS_NODELAY;
2304 rc = obd_statfs(NULL, sbi->ll_md_exp, osfs, max_age, flags);
2308 osfs->os_type = LL_SUPER_MAGIC;
2310 CDEBUG(D_SUPER, "MDC blocks %llu/%llu objects %llu/%llu\n",
2311 osfs->os_bavail, osfs->os_blocks, osfs->os_ffree, osfs->os_files);
2313 if (osfs->os_state & OS_STATFS_SUM)
2316 rc = obd_statfs(NULL, sbi->ll_dt_exp, &obd_osfs, max_age, flags);
2317 if (rc) /* Possibly a filesystem with no OSTs. Report MDT totals. */
2320 CDEBUG(D_SUPER, "OSC blocks %llu/%llu objects %llu/%llu\n",
2321 obd_osfs.os_bavail, obd_osfs.os_blocks, obd_osfs.os_ffree,
2324 osfs->os_bsize = obd_osfs.os_bsize;
2325 osfs->os_blocks = obd_osfs.os_blocks;
2326 osfs->os_bfree = obd_osfs.os_bfree;
2327 osfs->os_bavail = obd_osfs.os_bavail;
2329 /* If we have _some_ OSTs, but don't have as many free objects on the
2330 * OSTs as inodes on the MDTs, reduce the reported number of inodes
2331 * to compensate, so that the "inodes in use" number is correct.
2332 * This should be kept in sync with lod_statfs() behaviour.
2334 if (obd_osfs.os_files && obd_osfs.os_ffree < osfs->os_ffree) {
2335 osfs->os_files = (osfs->os_files - osfs->os_ffree) +
2337 osfs->os_ffree = obd_osfs.os_ffree;
2344 static int ll_statfs_project(struct inode *inode, struct kstatfs *sfs)
2346 struct if_quotactl qctl = {
2347 .qc_cmd = LUSTRE_Q_GETQUOTA,
2348 .qc_type = PRJQUOTA,
2349 .qc_valid = QC_GENERAL,
2351 u64 limit, curblock;
2354 qctl.qc_id = ll_i2info(inode)->lli_projid;
2355 ret = quotactl_ioctl(inode->i_sb, &qctl);
2357 /* ignore errors if project ID does not have
2358 * a quota limit or feature unsupported.
2360 if (ret == -ESRCH || ret == -EOPNOTSUPP)
2365 limit = ((qctl.qc_dqblk.dqb_bsoftlimit ?
2366 qctl.qc_dqblk.dqb_bsoftlimit :
2367 qctl.qc_dqblk.dqb_bhardlimit) * 1024) / sfs->f_bsize;
2368 if (limit && sfs->f_blocks > limit) {
2369 curblock = (qctl.qc_dqblk.dqb_curspace +
2370 sfs->f_bsize - 1) / sfs->f_bsize;
2371 sfs->f_blocks = limit;
2372 sfs->f_bfree = sfs->f_bavail =
2373 (sfs->f_blocks > curblock) ?
2374 (sfs->f_blocks - curblock) : 0;
2377 limit = qctl.qc_dqblk.dqb_isoftlimit ?
2378 qctl.qc_dqblk.dqb_isoftlimit :
2379 qctl.qc_dqblk.dqb_ihardlimit;
2380 if (limit && sfs->f_files > limit) {
2381 sfs->f_files = limit;
2382 sfs->f_ffree = (sfs->f_files >
2383 qctl.qc_dqblk.dqb_curinodes) ?
2384 (sfs->f_files - qctl.qc_dqblk.dqb_curinodes) : 0;
2390 int ll_statfs(struct dentry *de, struct kstatfs *sfs)
2392 struct super_block *sb = de->d_sb;
2393 struct obd_statfs osfs;
2394 __u64 fsid = huge_encode_dev(sb->s_dev);
2395 ktime_t kstart = ktime_get();
2398 CDEBUG(D_VFSTRACE, "VFS Op:sb=%s (%p)\n", sb->s_id, sb);
2400 /* Some amount of caching on the client is allowed */
2401 rc = ll_statfs_internal(ll_s2sbi(sb), &osfs, OBD_STATFS_SUM);
2405 statfs_unpack(sfs, &osfs);
2407 /* We need to downshift for all 32-bit kernels, because we can't
2408 * tell if the kernel is being called via sys_statfs64() or not.
2409 * Stop before overflowing f_bsize - in which case it is better
2410 * to just risk EOVERFLOW if caller is using old sys_statfs(). */
2411 if (sizeof(long) < 8) {
2412 while (osfs.os_blocks > ~0UL && sfs->f_bsize < 0x40000000) {
2415 osfs.os_blocks >>= 1;
2416 osfs.os_bfree >>= 1;
2417 osfs.os_bavail >>= 1;
2421 sfs->f_blocks = osfs.os_blocks;
2422 sfs->f_bfree = osfs.os_bfree;
2423 sfs->f_bavail = osfs.os_bavail;
2424 sfs->f_fsid.val[0] = (__u32)fsid;
2425 sfs->f_fsid.val[1] = (__u32)(fsid >> 32);
2426 if (ll_i2info(de->d_inode)->lli_projid)
2427 return ll_statfs_project(de->d_inode, sfs);
2429 ll_stats_ops_tally(ll_s2sbi(sb), LPROC_LL_STATFS,
2430 ktime_us_delta(ktime_get(), kstart));
2435 void ll_inode_size_lock(struct inode *inode)
2437 struct ll_inode_info *lli;
2439 LASSERT(!S_ISDIR(inode->i_mode));
2441 lli = ll_i2info(inode);
2442 mutex_lock(&lli->lli_size_mutex);
2445 void ll_inode_size_unlock(struct inode *inode)
2447 struct ll_inode_info *lli;
2449 lli = ll_i2info(inode);
2450 mutex_unlock(&lli->lli_size_mutex);
2453 void ll_update_inode_flags(struct inode *inode, unsigned int ext_flags)
2455 /* do not clear encryption flag */
2456 ext_flags |= ll_inode_to_ext_flags(inode->i_flags) & LUSTRE_ENCRYPT_FL;
2457 inode->i_flags = ll_ext_to_inode_flags(ext_flags);
2458 if (ext_flags & LUSTRE_PROJINHERIT_FL)
2459 set_bit(LLIF_PROJECT_INHERIT, &ll_i2info(inode)->lli_flags);
2461 clear_bit(LLIF_PROJECT_INHERIT, &ll_i2info(inode)->lli_flags);
2464 int ll_update_inode(struct inode *inode, struct lustre_md *md)
2466 struct ll_inode_info *lli = ll_i2info(inode);
2467 struct mdt_body *body = md->body;
2468 struct ll_sb_info *sbi = ll_i2sbi(inode);
2472 if (body->mbo_valid & OBD_MD_FLEASIZE) {
2473 rc = cl_file_inode_init(inode, md);
2478 if (S_ISDIR(inode->i_mode)) {
2479 rc = ll_update_lsm_md(inode, md);
2484 if (body->mbo_valid & OBD_MD_FLACL)
2485 lli_replace_acl(lli, md);
2487 api32 = test_bit(LL_SBI_32BIT_API, sbi->ll_flags);
2488 inode->i_ino = cl_fid_build_ino(&body->mbo_fid1, api32);
2489 inode->i_generation = cl_fid_build_gen(&body->mbo_fid1);
2491 if (body->mbo_valid & OBD_MD_FLATIME) {
2492 if (body->mbo_atime > inode->i_atime.tv_sec)
2493 inode->i_atime.tv_sec = body->mbo_atime;
2494 lli->lli_atime = body->mbo_atime;
2497 if (body->mbo_valid & OBD_MD_FLMTIME) {
2498 if (body->mbo_mtime > inode->i_mtime.tv_sec) {
2500 "setting ino %lu mtime from %lld to %llu\n",
2501 inode->i_ino, (s64)inode->i_mtime.tv_sec,
2503 inode->i_mtime.tv_sec = body->mbo_mtime;
2505 lli->lli_mtime = body->mbo_mtime;
2508 if (body->mbo_valid & OBD_MD_FLCTIME) {
2509 if (body->mbo_ctime > inode->i_ctime.tv_sec)
2510 inode->i_ctime.tv_sec = body->mbo_ctime;
2511 lli->lli_ctime = body->mbo_ctime;
2514 if (body->mbo_valid & OBD_MD_FLBTIME)
2515 lli->lli_btime = body->mbo_btime;
2517 /* Clear i_flags to remove S_NOSEC before permissions are updated */
2518 if (body->mbo_valid & OBD_MD_FLFLAGS)
2519 ll_update_inode_flags(inode, body->mbo_flags);
2520 if (body->mbo_valid & OBD_MD_FLMODE)
2521 inode->i_mode = (inode->i_mode & S_IFMT) |
2522 (body->mbo_mode & ~S_IFMT);
2524 if (body->mbo_valid & OBD_MD_FLTYPE)
2525 inode->i_mode = (inode->i_mode & ~S_IFMT) |
2526 (body->mbo_mode & S_IFMT);
2528 LASSERT(inode->i_mode != 0);
2529 if (body->mbo_valid & OBD_MD_FLUID)
2530 inode->i_uid = make_kuid(&init_user_ns, body->mbo_uid);
2531 if (body->mbo_valid & OBD_MD_FLGID)
2532 inode->i_gid = make_kgid(&init_user_ns, body->mbo_gid);
2533 if (body->mbo_valid & OBD_MD_FLPROJID)
2534 lli->lli_projid = body->mbo_projid;
2535 if (body->mbo_valid & OBD_MD_FLNLINK)
2536 set_nlink(inode, body->mbo_nlink);
2537 if (body->mbo_valid & OBD_MD_FLRDEV)
2538 inode->i_rdev = old_decode_dev(body->mbo_rdev);
2540 if (body->mbo_valid & OBD_MD_FLID) {
2541 /* FID shouldn't be changed! */
2542 if (fid_is_sane(&lli->lli_fid)) {
2543 LASSERTF(lu_fid_eq(&lli->lli_fid, &body->mbo_fid1),
2544 "Trying to change FID "DFID
2545 " to the "DFID", inode "DFID"(%p)\n",
2546 PFID(&lli->lli_fid), PFID(&body->mbo_fid1),
2547 PFID(ll_inode2fid(inode)), inode);
2549 lli->lli_fid = body->mbo_fid1;
2553 LASSERT(fid_seq(&lli->lli_fid) != 0);
2555 lli->lli_attr_valid = body->mbo_valid;
2556 if (body->mbo_valid & OBD_MD_FLSIZE) {
2557 i_size_write(inode, body->mbo_size);
2559 CDEBUG(D_VFSTRACE, "inode="DFID", updating i_size %llu\n",
2560 PFID(ll_inode2fid(inode)),
2561 (unsigned long long)body->mbo_size);
2563 if (body->mbo_valid & OBD_MD_FLBLOCKS)
2564 inode->i_blocks = body->mbo_blocks;
2566 if (body->mbo_valid & OBD_MD_FLLAZYSIZE)
2567 lli->lli_lazysize = body->mbo_size;
2568 if (body->mbo_valid & OBD_MD_FLLAZYBLOCKS)
2569 lli->lli_lazyblocks = body->mbo_blocks;
2572 if (body->mbo_valid & OBD_MD_TSTATE) {
2573 /* Set LLIF_FILE_RESTORING if restore ongoing and
2574 * clear it when done to ensure to start again
2575 * glimpsing updated attrs
2577 if (body->mbo_t_state & MS_RESTORE)
2578 set_bit(LLIF_FILE_RESTORING, &lli->lli_flags);
2580 clear_bit(LLIF_FILE_RESTORING, &lli->lli_flags);
2586 /* update directory depth to ROOT, called after LOOKUP lock is fetched. */
2587 void ll_update_dir_depth(struct inode *dir, struct inode *inode)
2589 struct ll_inode_info *lli;
2591 if (!S_ISDIR(inode->i_mode))
2597 lli = ll_i2info(inode);
2598 lli->lli_depth = ll_i2info(dir)->lli_depth + 1;
2599 CDEBUG(D_INODE, DFID" depth %hu\n", PFID(&lli->lli_fid), lli->lli_depth);
2602 void ll_truncate_inode_pages_final(struct inode *inode)
2604 struct address_space *mapping = &inode->i_data;
2605 unsigned long nrpages;
2606 unsigned long flags;
2608 truncate_inode_pages_final(mapping);
2610 /* Workaround for LU-118: Note nrpages may not be totally updated when
2611 * truncate_inode_pages() returns, as there can be a page in the process
2612 * of deletion (inside __delete_from_page_cache()) in the specified
2613 * range. Thus mapping->nrpages can be non-zero when this function
2614 * returns even after truncation of the whole mapping. Only do this if
2615 * npages isn't already zero.
2617 nrpages = mapping->nrpages;
2619 ll_xa_lock_irqsave(&mapping->i_pages, flags);
2620 nrpages = mapping->nrpages;
2621 ll_xa_unlock_irqrestore(&mapping->i_pages, flags);
2622 } /* Workaround end */
2624 LASSERTF(nrpages == 0, "%s: inode="DFID"(%p) nrpages=%lu, "
2625 "see https://jira.whamcloud.com/browse/LU-118\n",
2626 ll_i2sbi(inode)->ll_fsname,
2627 PFID(ll_inode2fid(inode)), inode, nrpages);
2630 int ll_read_inode2(struct inode *inode, void *opaque)
2632 struct lustre_md *md = opaque;
2633 struct ll_inode_info *lli = ll_i2info(inode);
2637 CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p)\n",
2638 PFID(&lli->lli_fid), inode);
2640 /* Core attributes from the MDS first. This is a new inode, and
2641 * the VFS doesn't zero times in the core inode so we have to do
2642 * it ourselves. They will be overwritten by either MDS or OST
2643 * attributes - we just need to make sure they aren't newer.
2645 inode->i_mtime.tv_sec = 0;
2646 inode->i_atime.tv_sec = 0;
2647 inode->i_ctime.tv_sec = 0;
2649 rc = ll_update_inode(inode, md);
2653 /* OIDEBUG(inode); */
2655 #ifdef HAVE_BACKING_DEV_INFO
2656 /* initializing backing dev info. */
2657 inode->i_mapping->backing_dev_info = &s2lsi(inode->i_sb)->lsi_bdi;
2659 if (S_ISREG(inode->i_mode)) {
2660 struct ll_sb_info *sbi = ll_i2sbi(inode);
2661 inode->i_op = &ll_file_inode_operations;
2662 inode->i_fop = sbi->ll_fop;
2663 inode->i_mapping->a_ops = (struct address_space_operations *)&ll_aops;
2665 } else if (S_ISDIR(inode->i_mode)) {
2666 inode->i_op = &ll_dir_inode_operations;
2667 inode->i_fop = &ll_dir_operations;
2669 } else if (S_ISLNK(inode->i_mode)) {
2670 inode->i_op = &ll_fast_symlink_inode_operations;
2673 inode->i_op = &ll_special_inode_operations;
2675 init_special_inode(inode, inode->i_mode,
2684 void ll_delete_inode(struct inode *inode)
2686 struct ll_inode_info *lli = ll_i2info(inode);
2689 if (S_ISREG(inode->i_mode) && lli->lli_clob != NULL) {
2690 /* It is last chance to write out dirty pages,
2691 * otherwise we may lose data while umount.
2693 * If i_nlink is 0 then just discard data. This is safe because
2694 * local inode gets i_nlink 0 from server only for the last
2695 * unlink, so that file is not opened somewhere else
2697 cl_sync_file_range(inode, 0, OBD_OBJECT_EOF, inode->i_nlink ?
2698 CL_FSYNC_LOCAL : CL_FSYNC_DISCARD, 1);
2701 ll_truncate_inode_pages_final(inode);
2702 ll_clear_inode(inode);
2708 int ll_iocontrol(struct inode *inode, struct file *file,
2709 unsigned int cmd, unsigned long arg)
2711 struct ll_sb_info *sbi = ll_i2sbi(inode);
2712 struct ptlrpc_request *req = NULL;
2717 case FS_IOC_GETFLAGS: {
2718 struct mdt_body *body;
2719 struct md_op_data *op_data;
2721 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL,
2722 0, 0, LUSTRE_OPC_ANY,
2724 if (IS_ERR(op_data))
2725 RETURN(PTR_ERR(op_data));
2727 op_data->op_valid = OBD_MD_FLFLAGS;
2728 rc = md_getattr(sbi->ll_md_exp, op_data, &req);
2729 ll_finish_md_op_data(op_data);
2731 CERROR("%s: failure inode "DFID": rc = %d\n",
2732 sbi->ll_md_exp->exp_obd->obd_name,
2733 PFID(ll_inode2fid(inode)), rc);
2737 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
2739 flags = body->mbo_flags;
2741 ptlrpc_req_finished(req);
2743 RETURN(put_user(flags, (int __user *)arg));
2745 case FS_IOC_SETFLAGS: {
2747 struct md_op_data *op_data;
2748 struct cl_object *obj;
2749 struct fsxattr fa = { 0 };
2751 if (get_user(flags, (int __user *)arg))
2754 fa.fsx_projid = ll_i2info(inode)->lli_projid;
2755 if (flags & LUSTRE_PROJINHERIT_FL)
2756 fa.fsx_xflags = FS_XFLAG_PROJINHERIT;
2758 rc = ll_ioctl_check_project(inode, fa.fsx_xflags,
2763 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
2764 LUSTRE_OPC_ANY, NULL);
2765 if (IS_ERR(op_data))
2766 RETURN(PTR_ERR(op_data));
2768 op_data->op_attr_flags = flags;
2769 op_data->op_xvalid |= OP_XVALID_FLAGS;
2770 rc = md_setattr(sbi->ll_md_exp, op_data, NULL, 0, &req);
2771 ll_finish_md_op_data(op_data);
2772 ptlrpc_req_finished(req);
2776 ll_update_inode_flags(inode, flags);
2778 obj = ll_i2info(inode)->lli_clob;
2782 OBD_ALLOC_PTR(attr);
2786 rc = cl_setattr_ost(obj, attr, OP_XVALID_FLAGS, flags);
2798 int ll_flush_ctx(struct inode *inode)
2800 struct ll_sb_info *sbi = ll_i2sbi(inode);
2802 CDEBUG(D_SEC, "flush context for user %d\n",
2803 from_kuid(&init_user_ns, current_uid()));
2805 obd_set_info_async(NULL, sbi->ll_md_exp,
2806 sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
2808 obd_set_info_async(NULL, sbi->ll_dt_exp,
2809 sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
2814 /* umount -f client means force down, don't save state */
2815 void ll_umount_begin(struct super_block *sb)
2817 struct ll_sb_info *sbi = ll_s2sbi(sb);
2818 struct obd_device *obd;
2819 struct obd_ioctl_data *ioc_data;
2823 CDEBUG(D_VFSTRACE, "VFS Op: superblock %p count %d active %d\n", sb,
2824 sb->s_count, atomic_read(&sb->s_active));
2826 obd = class_exp2obd(sbi->ll_md_exp);
2828 CERROR("Invalid MDC connection handle %#llx\n",
2829 sbi->ll_md_exp->exp_handle.h_cookie);
2835 obd = class_exp2obd(sbi->ll_dt_exp);
2837 CERROR("Invalid LOV connection handle %#llx\n",
2838 sbi->ll_dt_exp->exp_handle.h_cookie);
2844 OBD_ALLOC_PTR(ioc_data);
2846 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_md_exp,
2847 sizeof *ioc_data, ioc_data, NULL);
2849 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_dt_exp,
2850 sizeof *ioc_data, ioc_data, NULL);
2852 OBD_FREE_PTR(ioc_data);
2855 /* Really, we'd like to wait until there are no requests outstanding,
2856 * and then continue. For now, we just periodically checking for vfs
2857 * to decrement mnt_cnt and hope to finish it within 10sec.
2861 !may_umount(sbi->ll_mnt.mnt)) {
2869 int ll_remount_fs(struct super_block *sb, int *flags, char *data)
2871 struct ll_sb_info *sbi = ll_s2sbi(sb);
2872 char *profilenm = get_profile_name(sb);
2876 if ((*flags & MS_RDONLY) != (sb->s_flags & SB_RDONLY)) {
2877 read_only = *flags & MS_RDONLY;
2878 err = obd_set_info_async(NULL, sbi->ll_md_exp,
2879 sizeof(KEY_READ_ONLY),
2880 KEY_READ_ONLY, sizeof(read_only),
2883 LCONSOLE_WARN("Failed to remount %s %s (%d)\n",
2884 profilenm, read_only ?
2885 "read-only" : "read-write", err);
2890 sb->s_flags |= SB_RDONLY;
2892 sb->s_flags &= ~SB_RDONLY;
2894 if (test_bit(LL_SBI_VERBOSE, sbi->ll_flags))
2895 LCONSOLE_WARN("Remounted %s %s\n", profilenm,
2896 read_only ? "read-only" : "read-write");
2902 * Cleanup the open handle that is cached on MDT-side.
2904 * For open case, the client side open handling thread may hit error
2905 * after the MDT grant the open. Under such case, the client should
2906 * send close RPC to the MDT as cleanup; otherwise, the open handle
2907 * on the MDT will be leaked there until the client umount or evicted.
2909 * In further, if someone unlinked the file, because the open handle
2910 * holds the reference on such file/object, then it will block the
2911 * subsequent threads that want to locate such object via FID.
2913 * \param[in] sb super block for this file-system
2914 * \param[in] open_req pointer to the original open request
2916 void ll_open_cleanup(struct super_block *sb, struct req_capsule *pill)
2918 struct mdt_body *body;
2919 struct md_op_data *op_data;
2920 struct ptlrpc_request *close_req = NULL;
2921 struct obd_export *exp = ll_s2sbi(sb)->ll_md_exp;
2924 body = req_capsule_server_get(pill, &RMF_MDT_BODY);
2925 OBD_ALLOC_PTR(op_data);
2926 if (op_data == NULL) {
2927 CWARN("%s: cannot allocate op_data to release open handle for "
2928 DFID"\n", ll_s2sbi(sb)->ll_fsname, PFID(&body->mbo_fid1));
2933 op_data->op_fid1 = body->mbo_fid1;
2934 op_data->op_open_handle = body->mbo_open_handle;
2935 op_data->op_mod_time = ktime_get_real_seconds();
2936 md_close(exp, op_data, NULL, &close_req);
2937 ptlrpc_req_finished(close_req);
2938 ll_finish_md_op_data(op_data);
2943 int ll_prep_inode(struct inode **inode, struct req_capsule *pill,
2944 struct super_block *sb, struct lookup_intent *it)
2946 struct ll_sb_info *sbi = NULL;
2947 struct lustre_md md = { NULL };
2948 bool default_lmv_deleted = false;
2953 LASSERT(*inode || sb);
2954 sbi = sb ? ll_s2sbi(sb) : ll_i2sbi(*inode);
2955 rc = md_get_lustre_md(sbi->ll_md_exp, pill, sbi->ll_dt_exp,
2956 sbi->ll_md_exp, &md);
2961 * clear default_lmv only if intent_getattr reply doesn't contain it.
2962 * but it needs to be done after iget, check this early because
2963 * ll_update_lsm_md() may change md.
2965 if (it && (it->it_op & (IT_LOOKUP | IT_GETATTR)) &&
2966 S_ISDIR(md.body->mbo_mode) && !md.default_lmv)
2967 default_lmv_deleted = true;
2970 rc = ll_update_inode(*inode, &md);
2974 bool api32 = test_bit(LL_SBI_32BIT_API, sbi->ll_flags);
2975 struct lu_fid *fid1 = &md.body->mbo_fid1;
2977 LASSERT(sb != NULL);
2980 * At this point server returns to client's same fid as client
2981 * generated for creating. So using ->fid1 is okay here.
2983 if (!fid_is_sane(fid1)) {
2984 CERROR("%s: Fid is insane "DFID"\n",
2985 sbi->ll_fsname, PFID(fid1));
2986 GOTO(out, rc = -EINVAL);
2989 *inode = ll_iget(sb, cl_fid_build_ino(fid1, api32), &md);
2990 if (IS_ERR(*inode)) {
2992 rc = IS_ERR(*inode) ? PTR_ERR(*inode) : -ENOMEM;
2994 CERROR("new_inode -fatal: rc %d\n", rc);
2999 /* Handling piggyback layout lock.
3000 * Layout lock can be piggybacked by getattr and open request.
3001 * The lsm can be applied to inode only if it comes with a layout lock
3002 * otherwise correct layout may be overwritten, for example:
3003 * 1. proc1: mdt returns a lsm but not granting layout
3004 * 2. layout was changed by another client
3005 * 3. proc2: refresh layout and layout lock granted
3006 * 4. proc1: to apply a stale layout */
3007 if (it != NULL && it->it_lock_mode != 0) {
3008 struct lustre_handle lockh;
3009 struct ldlm_lock *lock;
3011 lockh.cookie = it->it_lock_handle;
3012 lock = ldlm_handle2lock(&lockh);
3013 LASSERT(lock != NULL);
3014 if (ldlm_has_layout(lock)) {
3015 struct cl_object_conf conf;
3017 memset(&conf, 0, sizeof(conf));
3018 conf.coc_opc = OBJECT_CONF_SET;
3019 conf.coc_inode = *inode;
3020 conf.coc_lock = lock;
3021 conf.u.coc_layout = md.layout;
3022 (void)ll_layout_conf(*inode, &conf);
3024 LDLM_LOCK_PUT(lock);
3027 if (default_lmv_deleted)
3028 ll_update_default_lsm_md(*inode, &md);
3030 /* we may want to apply some policy for foreign file/dir */
3031 if (ll_sbi_has_foreign_symlink(sbi)) {
3032 rc = ll_manage_foreign(*inode, &md);
3040 /* cleanup will be done if necessary */
3041 md_free_lustre_md(sbi->ll_md_exp, &md);
3043 if (rc != 0 && it != NULL && it->it_op & IT_OPEN) {
3044 ll_intent_drop_lock(it);
3045 ll_open_cleanup(sb != NULL ? sb : (*inode)->i_sb, pill);
3051 int ll_obd_statfs(struct inode *inode, void __user *arg)
3053 struct ll_sb_info *sbi = NULL;
3054 struct obd_export *exp;
3055 struct obd_ioctl_data *data = NULL;
3060 sbi = ll_i2sbi(inode);
3062 GOTO(out_statfs, rc = -EINVAL);
3064 rc = obd_ioctl_getdata(&data, &len, arg);
3066 GOTO(out_statfs, rc);
3068 if (!data->ioc_inlbuf1 || !data->ioc_inlbuf2 ||
3069 !data->ioc_pbuf1 || !data->ioc_pbuf2)
3070 GOTO(out_statfs, rc = -EINVAL);
3072 if (data->ioc_inllen1 != sizeof(__u32) ||
3073 data->ioc_inllen2 != sizeof(__u32) ||
3074 data->ioc_plen1 != sizeof(struct obd_statfs) ||
3075 data->ioc_plen2 != sizeof(struct obd_uuid))
3076 GOTO(out_statfs, rc = -EINVAL);
3078 memcpy(&type, data->ioc_inlbuf1, sizeof(__u32));
3079 if (type & LL_STATFS_LMV)
3080 exp = sbi->ll_md_exp;
3081 else if (type & LL_STATFS_LOV)
3082 exp = sbi->ll_dt_exp;
3084 GOTO(out_statfs, rc = -ENODEV);
3086 rc = obd_iocontrol(IOC_OBD_STATFS, exp, len, data, NULL);
3088 GOTO(out_statfs, rc);
3090 OBD_FREE_LARGE(data, len);
3095 * this is normally called in ll_fini_md_op_data(), but sometimes it needs to
3096 * be called early to avoid deadlock.
3098 void ll_unlock_md_op_lsm(struct md_op_data *op_data)
3100 if (op_data->op_mea2_sem) {
3101 up_read_non_owner(op_data->op_mea2_sem);
3102 op_data->op_mea2_sem = NULL;
3105 if (op_data->op_mea1_sem) {
3106 up_read_non_owner(op_data->op_mea1_sem);
3107 op_data->op_mea1_sem = NULL;
3111 /* this function prepares md_op_data hint for passing it down to MD stack. */
3112 struct md_op_data *ll_prep_md_op_data(struct md_op_data *op_data,
3113 struct inode *i1, struct inode *i2,
3114 const char *name, size_t namelen,
3115 __u32 mode, enum md_op_code opc,
3118 struct llcrypt_name fname = { 0 };
3121 LASSERT(i1 != NULL);
3124 /* Do not reuse namelen for something else. */
3126 return ERR_PTR(-EINVAL);
3128 if (namelen > ll_i2sbi(i1)->ll_namelen)
3129 return ERR_PTR(-ENAMETOOLONG);
3131 /* "/" is not valid name, but it's allowed */
3132 if (!lu_name_is_valid_2(name, namelen) &&
3133 strncmp("/", name, namelen) != 0)
3134 return ERR_PTR(-EINVAL);
3137 if (op_data == NULL)
3138 OBD_ALLOC_PTR(op_data);
3140 if (op_data == NULL)
3141 return ERR_PTR(-ENOMEM);
3143 ll_i2gids(op_data->op_suppgids, i1, i2);
3144 op_data->op_fid1 = *ll_inode2fid(i1);
3146 if (S_ISDIR(i1->i_mode)) {
3147 down_read_non_owner(&ll_i2info(i1)->lli_lsm_sem);
3148 op_data->op_mea1_sem = &ll_i2info(i1)->lli_lsm_sem;
3149 op_data->op_mea1 = ll_i2info(i1)->lli_lsm_md;
3150 op_data->op_default_mea1 = ll_i2info(i1)->lli_default_lsm_md;
3154 op_data->op_fid2 = *ll_inode2fid(i2);
3155 if (S_ISDIR(i2->i_mode)) {
3157 /* i2 is typically a child of i1, and MUST be
3158 * further from the root to avoid deadlocks.
3160 down_read_non_owner(&ll_i2info(i2)->lli_lsm_sem);
3161 op_data->op_mea2_sem =
3162 &ll_i2info(i2)->lli_lsm_sem;
3164 op_data->op_mea2 = ll_i2info(i2)->lli_lsm_md;
3167 fid_zero(&op_data->op_fid2);
3170 if (test_bit(LL_SBI_64BIT_HASH, ll_i2sbi(i1)->ll_flags))
3171 op_data->op_cli_flags |= CLI_HASH64;
3173 if (ll_need_32bit_api(ll_i2sbi(i1)))
3174 op_data->op_cli_flags |= CLI_API32;
3176 if (opc == LUSTRE_OPC_LOOKUP || opc == LUSTRE_OPC_CREATE) {
3177 /* In case of lookup, ll_setup_filename() has already been
3178 * called in ll_lookup_it(), so just take provided name.
3180 fname.disk_name.name = (unsigned char *)name;
3181 fname.disk_name.len = namelen;
3182 } else if (name && namelen) {
3183 struct qstr dname = QSTR_INIT(name, namelen);
3185 struct lu_fid *pfid = NULL;
3189 if (!S_ISDIR(i1->i_mode) && i2 && S_ISDIR(i2->i_mode)) {
3190 /* special case when called from ll_link() */
3195 lookup = (int)(opc == LUSTRE_OPC_ANY);
3197 if (opc == LUSTRE_OPC_ANY && lookup)
3199 rc = ll_setup_filename(dir, &dname, lookup, &fname, pfid);
3201 ll_finish_md_op_data(op_data);
3204 if (pfid && !fid_is_zero(pfid)) {
3206 op_data->op_fid2 = fid;
3207 op_data->op_bias = MDS_FID_OP;
3209 if (fname.disk_name.name &&
3210 fname.disk_name.name != (unsigned char *)name)
3211 /* op_data->op_name must be freed after use */
3212 op_data->op_flags |= MF_OPNAME_KMALLOCED;
3215 /* In fact LUSTRE_OPC_LOOKUP, LUSTRE_OPC_OPEN, LUSTRE_OPC_MIGR
3216 * are LUSTRE_OPC_ANY
3218 if (opc == LUSTRE_OPC_LOOKUP || opc == LUSTRE_OPC_OPEN ||
3219 opc == LUSTRE_OPC_MIGR)
3220 op_data->op_code = LUSTRE_OPC_ANY;
3222 op_data->op_code = opc;
3223 op_data->op_name = fname.disk_name.name;
3224 op_data->op_namelen = fname.disk_name.len;
3225 op_data->op_mode = mode;
3226 op_data->op_mod_time = ktime_get_real_seconds();
3227 op_data->op_fsuid = from_kuid(&init_user_ns, current_fsuid());
3228 op_data->op_fsgid = from_kgid(&init_user_ns, current_fsgid());
3229 op_data->op_cap = current_cap();
3230 op_data->op_mds = 0;
3231 if ((opc == LUSTRE_OPC_CREATE) && (name != NULL) &&
3232 filename_is_volatile(name, namelen, &op_data->op_mds)) {
3233 op_data->op_bias |= MDS_CREATE_VOLATILE;
3235 op_data->op_data = data;
3240 void ll_finish_md_op_data(struct md_op_data *op_data)
3242 ll_unlock_md_op_lsm(op_data);
3243 ll_security_release_secctx(op_data->op_file_secctx,
3244 op_data->op_file_secctx_size);
3245 if (op_data->op_flags & MF_OPNAME_KMALLOCED)
3246 /* allocated via ll_setup_filename called
3247 * from ll_prep_md_op_data
3249 kfree(op_data->op_name);
3250 llcrypt_free_ctx(op_data->op_file_encctx, op_data->op_file_encctx_size);
3251 OBD_FREE_PTR(op_data);
3254 int ll_show_options(struct seq_file *seq, struct dentry *dentry)
3256 struct ll_sb_info *sbi;
3259 LASSERT(seq && dentry);
3260 sbi = ll_s2sbi(dentry->d_sb);
3262 if (test_bit(LL_SBI_NOLCK, sbi->ll_flags))
3263 seq_puts(seq, "nolock");
3265 for (i = 1; ll_sbi_flags_name[i].token != LL_SBI_NUM_MOUNT_OPT; i++) {
3266 /* match_table in some cases has patterns for both enabled and
3267 * disabled cases. Ignore 'no'xxx versions if bit is set.
3269 if (test_bit(ll_sbi_flags_name[i].token, sbi->ll_flags) &&
3270 strncmp(ll_sbi_flags_name[i].pattern, "no", 2)) {
3271 if (ll_sbi_flags_name[i].token ==
3272 LL_SBI_FOREIGN_SYMLINK) {
3273 seq_show_option(seq, "foreign_symlink",
3274 sbi->ll_foreign_symlink_prefix);
3276 seq_printf(seq, ",%s",
3277 ll_sbi_flags_name[i].pattern);
3280 /* You can have either localflock or flock but not
3281 * both. If localflock is set don't print flock or
3284 if (ll_sbi_flags_name[i].token == LL_SBI_LOCALFLOCK)
3286 } else if (!test_bit(ll_sbi_flags_name[i].token, sbi->ll_flags) &&
3287 !strncmp(ll_sbi_flags_name[i].pattern, "no", 2)) {
3288 seq_printf(seq, ",%s",
3289 ll_sbi_flags_name[i].pattern);
3297 * Get obd name by cmd, and copy out to user space
3299 int ll_get_obd_name(struct inode *inode, unsigned int cmd, unsigned long arg)
3301 struct ll_sb_info *sbi = ll_i2sbi(inode);
3302 struct obd_device *obd;
3305 if (cmd == OBD_IOC_GETNAME_OLD || cmd == OBD_IOC_GETDTNAME)
3306 obd = class_exp2obd(sbi->ll_dt_exp);
3307 else if (cmd == OBD_IOC_GETMDNAME)
3308 obd = class_exp2obd(sbi->ll_md_exp);
3315 if (copy_to_user((void __user *)arg, obd->obd_name,
3316 strlen(obd->obd_name) + 1))
3322 static char* ll_d_path(struct dentry *dentry, char *buf, int bufsize)
3329 p.mnt = current->fs->root.mnt;
3331 path = d_path(&p, buf, bufsize);
3336 void ll_dirty_page_discard_warn(struct page *page, int ioret)
3338 char *buf, *path = NULL;
3339 struct dentry *dentry = NULL;
3340 struct inode *inode = page->mapping->host;
3342 /* this can be called inside spin lock so use GFP_ATOMIC. */
3343 buf = (char *)__get_free_page(GFP_ATOMIC);
3345 dentry = d_find_alias(page->mapping->host);
3347 path = ll_d_path(dentry, buf, PAGE_SIZE);
3350 /* The below message is checked in recovery-small.sh test_24b */
3352 "%s: dirty page discard: %s/fid: "DFID"/%s may get corrupted "
3353 "(rc %d)\n", ll_i2sbi(inode)->ll_fsname,
3354 s2lsi(page->mapping->host->i_sb)->lsi_lmd->lmd_dev,
3355 PFID(ll_inode2fid(inode)),
3356 (path && !IS_ERR(path)) ? path : "", ioret);
3362 free_page((unsigned long)buf);
3365 ssize_t ll_copy_user_md(const struct lov_user_md __user *md,
3366 struct lov_user_md **kbuf)
3368 struct lov_user_md lum;
3372 if (copy_from_user(&lum, md, sizeof(lum)))
3375 lum_size = ll_lov_user_md_size(&lum);
3379 OBD_ALLOC_LARGE(*kbuf, lum_size);
3383 if (copy_from_user(*kbuf, md, lum_size) != 0) {
3384 OBD_FREE_LARGE(*kbuf, lum_size);
3392 * Compute llite root squash state after a change of root squash
3393 * configuration setting or add/remove of a lnet nid
3395 void ll_compute_rootsquash_state(struct ll_sb_info *sbi)
3397 struct root_squash_info *squash = &sbi->ll_squash;
3400 struct lnet_process_id id;
3402 /* Update norootsquash flag */
3403 spin_lock(&squash->rsi_lock);
3404 if (list_empty(&squash->rsi_nosquash_nids))
3405 clear_bit(LL_SBI_NOROOTSQUASH, sbi->ll_flags);
3407 /* Do not apply root squash as soon as one of our NIDs is
3408 * in the nosquash_nids list */
3411 while (LNetGetId(i++, &id) != -ENOENT) {
3412 if (id.nid == LNET_NID_LO_0)
3414 if (cfs_match_nid(id.nid, &squash->rsi_nosquash_nids)) {
3420 set_bit(LL_SBI_NOROOTSQUASH, sbi->ll_flags);
3422 clear_bit(LL_SBI_NOROOTSQUASH, sbi->ll_flags);
3424 spin_unlock(&squash->rsi_lock);
3428 * Parse linkea content to extract information about a given hardlink
3430 * \param[in] ldata - Initialized linkea data
3431 * \param[in] linkno - Link identifier
3432 * \param[out] parent_fid - The entry's parent FID
3433 * \param[out] ln - Entry name destination buffer
3435 * \retval 0 on success
3436 * \retval Appropriate negative error code on failure
3438 static int ll_linkea_decode(struct linkea_data *ldata, unsigned int linkno,
3439 struct lu_fid *parent_fid, struct lu_name *ln)
3445 rc = linkea_init_with_rec(ldata);
3449 if (linkno >= ldata->ld_leh->leh_reccount)
3450 /* beyond last link */
3453 linkea_first_entry(ldata);
3454 for (idx = 0; ldata->ld_lee != NULL; idx++) {
3455 linkea_entry_unpack(ldata->ld_lee, &ldata->ld_reclen, ln,
3460 linkea_next_entry(ldata);
3470 * Get parent FID and name of an identified link. Operation is performed for
3471 * a given link number, letting the caller iterate over linkno to list one or
3472 * all links of an entry.
3474 * \param[in] file - File descriptor against which to perform the operation
3475 * \param[in,out] arg - User-filled structure containing the linkno to operate
3476 * on and the available size. It is eventually filled with
3477 * the requested information or left untouched on error
3479 * \retval - 0 on success
3480 * \retval - Appropriate negative error code on failure
3482 int ll_getparent(struct file *file, struct getparent __user *arg)
3484 struct inode *inode = file_inode(file);
3485 struct linkea_data *ldata;
3486 struct lu_buf buf = LU_BUF_NULL;
3488 struct lu_fid parent_fid;
3495 if (!capable(CAP_DAC_READ_SEARCH) &&
3496 !test_bit(LL_SBI_USER_FID2PATH, ll_i2sbi(inode)->ll_flags))
3499 if (get_user(name_size, &arg->gp_name_size))
3502 if (get_user(linkno, &arg->gp_linkno))
3505 if (name_size > PATH_MAX)
3508 OBD_ALLOC(ldata, sizeof(*ldata));
3512 rc = linkea_data_new(ldata, &buf);
3514 GOTO(ldata_free, rc);
3516 rc = ll_xattr_list(inode, XATTR_NAME_LINK, XATTR_TRUSTED_T, buf.lb_buf,
3517 buf.lb_len, OBD_MD_FLXATTR);
3521 rc = ll_linkea_decode(ldata, linkno, &parent_fid, &ln);
3525 if (ln.ln_namelen >= name_size)
3526 GOTO(lb_free, rc = -EOVERFLOW);
3528 if (copy_to_user(&arg->gp_fid, &parent_fid, sizeof(arg->gp_fid)))
3529 GOTO(lb_free, rc = -EFAULT);
3531 if (copy_to_user(&arg->gp_name, ln.ln_name, ln.ln_namelen))
3532 GOTO(lb_free, rc = -EFAULT);
3534 if (put_user('\0', arg->gp_name + ln.ln_namelen))
3535 GOTO(lb_free, rc = -EFAULT);
3540 OBD_FREE(ldata, sizeof(*ldata));