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_GRANT_SHRINK |
328 OBD_CONNECT_SHORTIO | OBD_CONNECT_FLAGS2;
330 data->ocd_connect_flags2 = OBD_CONNECT2_DIR_MIGRATE |
331 OBD_CONNECT2_SUM_STATFS |
332 OBD_CONNECT2_OVERSTRIPING |
334 OBD_CONNECT2_LOCK_CONVERT |
335 OBD_CONNECT2_ARCHIVE_ID_ARRAY |
336 OBD_CONNECT2_INC_XID |
338 OBD_CONNECT2_ASYNC_DISCARD |
340 OBD_CONNECT2_CRUSH | OBD_CONNECT2_LSEEK |
341 OBD_CONNECT2_GETATTR_PFID |
342 OBD_CONNECT2_DOM_LVB |
343 OBD_CONNECT2_REP_MBITS |
344 OBD_CONNECT2_ATOMIC_OPEN_LOCK;
346 #ifdef HAVE_LRU_RESIZE_SUPPORT
347 if (test_bit(LL_SBI_LRU_RESIZE, sbi->ll_flags))
348 data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
350 data->ocd_connect_flags |= OBD_CONNECT_ACL_FLAGS;
352 data->ocd_cksum_types = obd_cksum_types_supported_client();
354 if (OBD_FAIL_CHECK(OBD_FAIL_MDC_LIGHTWEIGHT))
355 /* flag mdc connection as lightweight, only used for test
356 * purpose, use with care */
357 data->ocd_connect_flags |= OBD_CONNECT_LIGHTWEIGHT;
359 data->ocd_ibits_known = MDS_INODELOCK_FULL;
360 data->ocd_version = LUSTRE_VERSION_CODE;
362 if (sb->s_flags & SB_RDONLY)
363 data->ocd_connect_flags |= OBD_CONNECT_RDONLY;
364 if (test_bit(LL_SBI_USER_XATTR, sbi->ll_flags))
365 data->ocd_connect_flags |= OBD_CONNECT_XATTR;
368 /* Setting this indicates we correctly support S_NOSEC (See kernel
369 * commit 9e1f1de02c2275d7172e18dc4e7c2065777611bf)
371 sb->s_flags |= SB_NOSEC;
373 sbi->ll_fop = ll_select_file_operations(sbi);
375 /* always ping even if server suppress_pings */
376 if (test_bit(LL_SBI_ALWAYS_PING, sbi->ll_flags))
377 data->ocd_connect_flags &= ~OBD_CONNECT_PINGLESS;
379 obd_connect_set_secctx(data);
380 if (ll_sbi_has_encrypt(sbi))
381 obd_connect_set_enc(data);
383 #if defined(CONFIG_SECURITY)
384 data->ocd_connect_flags2 |= OBD_CONNECT2_SELINUX_POLICY;
387 data->ocd_brw_size = MD_MAX_BRW_SIZE;
389 err = obd_connect(NULL, &sbi->ll_md_exp, sbi->ll_md_obd,
390 &sbi->ll_sb_uuid, data, sbi->ll_cache);
392 LCONSOLE_ERROR_MSG(0x14f, "An MDT (md %s) is performing "
393 "recovery, of which this client is not a "
394 "part. Please wait for recovery to complete,"
395 " abort, or time out.\n", md);
398 CERROR("cannot connect to %s: rc = %d\n", md, err);
402 sbi->ll_md_exp->exp_connect_data = *data;
404 err = obd_fid_init(sbi->ll_md_exp->exp_obd, sbi->ll_md_exp,
405 LUSTRE_SEQ_METADATA);
407 CERROR("%s: Can't init metadata layer FID infrastructure, "
408 "rc = %d\n", sbi->ll_md_exp->exp_obd->obd_name, err);
412 /* For mount, we only need fs info from MDT0, and also in DNE, it
413 * can make sure the client can be mounted as long as MDT0 is
415 err = obd_statfs(NULL, sbi->ll_md_exp, osfs,
416 ktime_get_seconds() - sbi->ll_statfs_max_age,
417 OBD_STATFS_FOR_MDT0);
419 GOTO(out_md_fid, err);
421 /* This needs to be after statfs to ensure connect has finished.
422 * Note that "data" does NOT contain the valid connect reply.
423 * If connecting to a 1.8 server there will be no LMV device, so
424 * we can access the MDC export directly and exp_connect_flags will
425 * be non-zero, but if accessing an upgraded 2.1 server it will
426 * have the correct flags filled in.
427 * XXX: fill in the LMV exp_connect_flags from MDC(s). */
428 valid = exp_connect_flags(sbi->ll_md_exp) & CLIENT_CONNECT_MDT_REQD;
429 if (exp_connect_flags(sbi->ll_md_exp) != 0 &&
430 valid != CLIENT_CONNECT_MDT_REQD) {
433 OBD_ALLOC_WAIT(buf, PAGE_SIZE);
434 obd_connect_flags2str(buf, PAGE_SIZE,
435 valid ^ CLIENT_CONNECT_MDT_REQD, 0, ",");
436 LCONSOLE_ERROR_MSG(0x170, "Server %s does not support "
437 "feature(s) needed for correct operation "
438 "of this client (%s). Please upgrade "
439 "server or downgrade client.\n",
440 sbi->ll_md_exp->exp_obd->obd_name, buf);
441 OBD_FREE(buf, PAGE_SIZE);
442 GOTO(out_md_fid, err = -EPROTO);
445 size = sizeof(*data);
446 err = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_CONN_DATA),
447 KEY_CONN_DATA, &size, data);
449 CERROR("%s: Get connect data failed: rc = %d\n",
450 sbi->ll_md_exp->exp_obd->obd_name, err);
451 GOTO(out_md_fid, err);
454 LASSERT(osfs->os_bsize);
455 sb->s_blocksize = osfs->os_bsize;
456 sb->s_blocksize_bits = log2(osfs->os_bsize);
457 sb->s_magic = LL_SUPER_MAGIC;
458 sb->s_maxbytes = MAX_LFS_FILESIZE;
459 sbi->ll_namelen = osfs->os_namelen;
460 sbi->ll_mnt.mnt = current->fs->root.mnt;
462 if (test_bit(LL_SBI_USER_XATTR, sbi->ll_flags) &&
463 !(data->ocd_connect_flags & OBD_CONNECT_XATTR)) {
464 LCONSOLE_INFO("Disabling user_xattr feature because "
465 "it is not supported on the server\n");
466 clear_bit(LL_SBI_USER_XATTR, sbi->ll_flags);
469 if (data->ocd_connect_flags & OBD_CONNECT_ACL) {
471 sb->s_flags |= SB_POSIXACL;
473 set_bit(LL_SBI_ACL, sbi->ll_flags);
475 LCONSOLE_INFO("client wants to enable acl, but mdt not!\n");
477 sb->s_flags &= ~SB_POSIXACL;
479 clear_bit(LL_SBI_ACL, sbi->ll_flags);
482 if (data->ocd_connect_flags & OBD_CONNECT_64BITHASH)
483 set_bit(LL_SBI_64BIT_HASH, sbi->ll_flags);
485 if (data->ocd_connect_flags & OBD_CONNECT_LAYOUTLOCK)
486 set_bit(LL_SBI_LAYOUT_LOCK, sbi->ll_flags);
488 if (obd_connect_has_secctx(data))
489 set_bit(LL_SBI_FILE_SECCTX, sbi->ll_flags);
491 if (ll_sbi_has_encrypt(sbi) && !obd_connect_has_enc(data)) {
492 if (ll_sbi_has_test_dummy_encryption(sbi))
493 LCONSOLE_WARN("%s: server %s does not support encryption feature, encryption deactivated.\n",
495 sbi->ll_md_exp->exp_obd->obd_name);
496 ll_sbi_set_encrypt(sbi, false);
499 if (data->ocd_ibits_known & MDS_INODELOCK_XATTR) {
500 if (!(data->ocd_connect_flags & OBD_CONNECT_MAX_EASIZE)) {
501 LCONSOLE_INFO("%s: disabling xattr cache due to "
502 "unknown maximum xattr size.\n", dt);
503 } else if (!sbi->ll_xattr_cache_set) {
504 /* If xattr_cache is already set (no matter 0 or 1)
505 * during processing llog, it won't be enabled here. */
506 set_bit(LL_SBI_XATTR_CACHE, sbi->ll_flags);
507 sbi->ll_xattr_cache_enabled = 1;
511 sbi->ll_dt_obd = class_name2obd(dt);
512 if (!sbi->ll_dt_obd) {
513 CERROR("DT %s: not setup or attached\n", dt);
514 GOTO(out_md_fid, err = -ENODEV);
517 /* pass client page size via ocd_grant_blkbits, the server should report
518 * back its backend blocksize for grant calculation purpose */
519 data->ocd_grant_blkbits = PAGE_SHIFT;
521 /* indicate OST features supported by this client */
522 data->ocd_connect_flags = OBD_CONNECT_GRANT | OBD_CONNECT_VERSION |
523 OBD_CONNECT_REQPORTAL | OBD_CONNECT_BRW_SIZE |
524 OBD_CONNECT_CANCELSET | OBD_CONNECT_FID |
525 OBD_CONNECT_SRVLOCK |
526 OBD_CONNECT_AT | OBD_CONNECT_OSS_CAPA |
527 OBD_CONNECT_VBR | OBD_CONNECT_FULL20 |
528 OBD_CONNECT_64BITHASH | OBD_CONNECT_MAXBYTES |
529 OBD_CONNECT_EINPROGRESS |
530 OBD_CONNECT_JOBSTATS | OBD_CONNECT_LVB_TYPE |
531 OBD_CONNECT_LAYOUTLOCK |
532 OBD_CONNECT_PINGLESS | OBD_CONNECT_LFSCK |
533 OBD_CONNECT_BULK_MBITS | OBD_CONNECT_SHORTIO |
534 OBD_CONNECT_FLAGS2 | OBD_CONNECT_GRANT_SHRINK;
535 data->ocd_connect_flags2 = OBD_CONNECT2_LOCKAHEAD |
536 OBD_CONNECT2_INC_XID | OBD_CONNECT2_LSEEK |
537 OBD_CONNECT2_REP_MBITS;
539 if (!OBD_FAIL_CHECK(OBD_FAIL_OSC_CONNECT_GRANT_PARAM))
540 data->ocd_connect_flags |= OBD_CONNECT_GRANT_PARAM;
542 /* OBD_CONNECT_CKSUM should always be set, even if checksums are
543 * disabled by default, because it can still be enabled on the
544 * fly via /sys. As a consequence, we still need to come to an
545 * agreement on the supported algorithms at connect time
547 data->ocd_connect_flags |= OBD_CONNECT_CKSUM;
549 if (OBD_FAIL_CHECK(OBD_FAIL_OSC_CKSUM_ADLER_ONLY))
550 data->ocd_cksum_types = OBD_CKSUM_ADLER;
552 data->ocd_cksum_types = obd_cksum_types_supported_client();
554 #ifdef HAVE_LRU_RESIZE_SUPPORT
555 data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
557 /* always ping even if server suppress_pings */
558 if (test_bit(LL_SBI_ALWAYS_PING, sbi->ll_flags))
559 data->ocd_connect_flags &= ~OBD_CONNECT_PINGLESS;
561 if (ll_sbi_has_encrypt(sbi))
562 obd_connect_set_enc(data);
564 CDEBUG(D_RPCTRACE, "ocd_connect_flags: %#llx ocd_version: %d "
565 "ocd_grant: %d\n", data->ocd_connect_flags,
566 data->ocd_version, data->ocd_grant);
568 sbi->ll_dt_obd->obd_upcall.onu_owner = &sbi->ll_lco;
569 sbi->ll_dt_obd->obd_upcall.onu_upcall = cl_ocd_update;
571 data->ocd_brw_size = DT_MAX_BRW_SIZE;
573 err = obd_connect(NULL, &sbi->ll_dt_exp, sbi->ll_dt_obd,
574 &sbi->ll_sb_uuid, data, sbi->ll_cache);
576 LCONSOLE_ERROR_MSG(0x150, "An OST (dt %s) is performing "
577 "recovery, of which this client is not a "
578 "part. Please wait for recovery to "
579 "complete, abort, or time out.\n", dt);
582 CERROR("%s: Cannot connect to %s: rc = %d\n",
583 sbi->ll_dt_exp->exp_obd->obd_name, dt, err);
587 if (ll_sbi_has_encrypt(sbi) &&
588 !obd_connect_has_enc(&sbi->ll_dt_obd->u.lov.lov_ocd)) {
589 if (ll_sbi_has_test_dummy_encryption(sbi))
590 LCONSOLE_WARN("%s: server %s does not support encryption feature, encryption deactivated.\n",
592 ll_sbi_set_encrypt(sbi, false);
593 } else if (ll_sbi_has_test_dummy_encryption(sbi)) {
594 LCONSOLE_WARN("Test dummy encryption mode enabled\n");
597 sbi->ll_dt_exp->exp_connect_data = *data;
599 /* Don't change value if it was specified in the config log */
600 if (sbi->ll_ra_info.ra_max_read_ahead_whole_pages == -1) {
601 sbi->ll_ra_info.ra_max_read_ahead_whole_pages =
602 max_t(unsigned long, SBI_DEFAULT_READ_AHEAD_WHOLE_MAX,
603 (data->ocd_brw_size >> PAGE_SHIFT));
604 if (sbi->ll_ra_info.ra_max_read_ahead_whole_pages >
605 sbi->ll_ra_info.ra_max_pages_per_file)
606 sbi->ll_ra_info.ra_max_read_ahead_whole_pages =
607 sbi->ll_ra_info.ra_max_pages_per_file;
610 err = obd_fid_init(sbi->ll_dt_exp->exp_obd, sbi->ll_dt_exp,
611 LUSTRE_SEQ_METADATA);
613 CERROR("%s: Can't init data layer FID infrastructure, "
614 "rc = %d\n", sbi->ll_dt_exp->exp_obd->obd_name, err);
618 mutex_lock(&sbi->ll_lco.lco_lock);
619 sbi->ll_lco.lco_flags = data->ocd_connect_flags;
620 sbi->ll_lco.lco_md_exp = sbi->ll_md_exp;
621 sbi->ll_lco.lco_dt_exp = sbi->ll_dt_exp;
622 mutex_unlock(&sbi->ll_lco.lco_lock);
624 fid_zero(&sbi->ll_root_fid);
625 err = md_get_root(sbi->ll_md_exp, get_mount_fileset(sb),
628 CERROR("cannot mds_connect: rc = %d\n", err);
629 GOTO(out_lock_cn_cb, err);
631 if (!fid_is_sane(&sbi->ll_root_fid)) {
632 CERROR("%s: Invalid root fid "DFID" during mount\n",
633 sbi->ll_md_exp->exp_obd->obd_name,
634 PFID(&sbi->ll_root_fid));
635 GOTO(out_lock_cn_cb, err = -EINVAL);
637 CDEBUG(D_SUPER, "rootfid "DFID"\n", PFID(&sbi->ll_root_fid));
639 sb->s_op = &lustre_super_operations;
640 sb->s_xattr = ll_xattr_handlers;
641 #if THREAD_SIZE >= 8192 /*b=17630*/
642 sb->s_export_op = &lustre_export_operations;
644 #ifdef HAVE_LUSTRE_CRYPTO
645 llcrypt_set_ops(sb, &lustre_cryptops);
649 * XXX: move this to after cbd setup? */
650 valid = OBD_MD_FLGETATTR | OBD_MD_FLBLOCKS | OBD_MD_FLMODEASIZE;
651 if (test_bit(LL_SBI_ACL, sbi->ll_flags))
652 valid |= OBD_MD_FLACL;
654 OBD_ALLOC_PTR(op_data);
656 GOTO(out_lock_cn_cb, err = -ENOMEM);
658 op_data->op_fid1 = sbi->ll_root_fid;
659 op_data->op_mode = 0;
660 op_data->op_valid = valid;
662 err = md_getattr(sbi->ll_md_exp, op_data, &request);
664 OBD_FREE_PTR(op_data);
666 CERROR("%s: md_getattr failed for root: rc = %d\n",
667 sbi->ll_md_exp->exp_obd->obd_name, err);
668 GOTO(out_lock_cn_cb, err);
671 err = md_get_lustre_md(sbi->ll_md_exp, &request->rq_pill,
672 sbi->ll_dt_exp, sbi->ll_md_exp, &lmd);
674 CERROR("failed to understand root inode md: rc = %d\n", err);
675 ptlrpc_req_finished(request);
676 GOTO(out_lock_cn_cb, err);
679 LASSERT(fid_is_sane(&sbi->ll_root_fid));
680 api32 = test_bit(LL_SBI_32BIT_API, sbi->ll_flags);
681 root = ll_iget(sb, cl_fid_build_ino(&sbi->ll_root_fid, api32), &lmd);
682 md_free_lustre_md(sbi->ll_md_exp, &lmd);
683 ptlrpc_req_finished(request);
687 err = IS_ERR(root) ? PTR_ERR(root) : -EBADF;
689 CERROR("%s: bad ll_iget() for root: rc = %d\n",
690 sbi->ll_fsname, err);
694 checksum = test_bit(LL_SBI_CHECKSUM, sbi->ll_flags);
695 if (sbi->ll_checksum_set) {
696 err = obd_set_info_async(NULL, sbi->ll_dt_exp,
697 sizeof(KEY_CHECKSUM), KEY_CHECKSUM,
698 sizeof(checksum), &checksum, NULL);
700 CERROR("%s: Set checksum failed: rc = %d\n",
701 sbi->ll_dt_exp->exp_obd->obd_name, err);
707 sb->s_root = d_make_root(root);
708 if (sb->s_root == NULL) {
710 CERROR("%s: can't make root dentry: rc = %d\n",
711 sbi->ll_fsname, err);
715 sbi->ll_sdev_orig = sb->s_dev;
717 /* We set sb->s_dev equal on all lustre clients in order to support
718 * NFS export clustering. NFSD requires that the FSID be the same
720 /* s_dev is also used in lt_compare() to compare two fs, but that is
721 * only a node-local comparison. */
722 uuid = obd_get_uuid(sbi->ll_md_exp);
724 sb->s_dev = get_uuid2int(uuid->uuid, strlen(uuid->uuid));
731 if (sbi->ll_dt_obd) {
732 err = sysfs_create_link(&sbi->ll_kset.kobj,
733 &sbi->ll_dt_obd->obd_kset.kobj,
734 sbi->ll_dt_obd->obd_type->typ_name);
736 CERROR("%s: could not register %s in llite: rc = %d\n",
737 dt, sbi->ll_fsname, err);
742 if (sbi->ll_md_obd) {
743 err = sysfs_create_link(&sbi->ll_kset.kobj,
744 &sbi->ll_md_obd->obd_kset.kobj,
745 sbi->ll_md_obd->obd_type->typ_name);
747 CERROR("%s: could not register %s in llite: rc = %d\n",
748 md, sbi->ll_fsname, err);
757 obd_fid_fini(sbi->ll_dt_exp->exp_obd);
759 obd_disconnect(sbi->ll_dt_exp);
760 sbi->ll_dt_exp = NULL;
761 sbi->ll_dt_obd = NULL;
763 obd_fid_fini(sbi->ll_md_exp->exp_obd);
765 obd_disconnect(sbi->ll_md_exp);
766 sbi->ll_md_exp = NULL;
767 sbi->ll_md_obd = NULL;
776 int ll_get_max_mdsize(struct ll_sb_info *sbi, int *lmmsize)
780 size = sizeof(*lmmsize);
781 rc = obd_get_info(NULL, sbi->ll_dt_exp, sizeof(KEY_MAX_EASIZE),
782 KEY_MAX_EASIZE, &size, lmmsize);
784 CERROR("%s: cannot get max LOV EA size: rc = %d\n",
785 sbi->ll_dt_exp->exp_obd->obd_name, rc);
789 CDEBUG(D_INFO, "max LOV ea size: %d\n", *lmmsize);
792 rc = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_MAX_EASIZE),
793 KEY_MAX_EASIZE, &size, lmmsize);
795 CERROR("Get max mdsize error rc %d\n", rc);
797 CDEBUG(D_INFO, "max LMV ea size: %d\n", *lmmsize);
803 * Get the value of the default_easize parameter.
805 * \see client_obd::cl_default_mds_easize
807 * \param[in] sbi superblock info for this filesystem
808 * \param[out] lmmsize pointer to storage location for value
810 * \retval 0 on success
811 * \retval negative negated errno on failure
813 int ll_get_default_mdsize(struct ll_sb_info *sbi, int *lmmsize)
818 rc = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_DEFAULT_EASIZE),
819 KEY_DEFAULT_EASIZE, &size, lmmsize);
821 CERROR("Get default mdsize error rc %d\n", rc);
827 * Set the default_easize parameter to the given value.
829 * \see client_obd::cl_default_mds_easize
831 * \param[in] sbi superblock info for this filesystem
832 * \param[in] lmmsize the size to set
834 * \retval 0 on success
835 * \retval negative negated errno on failure
837 int ll_set_default_mdsize(struct ll_sb_info *sbi, int lmmsize)
841 if (lmmsize < sizeof(struct lov_mds_md) ||
842 lmmsize > OBD_MAX_DEFAULT_EA_SIZE)
845 rc = obd_set_info_async(NULL, sbi->ll_md_exp,
846 sizeof(KEY_DEFAULT_EASIZE), KEY_DEFAULT_EASIZE,
847 sizeof(int), &lmmsize, NULL);
852 static void client_common_put_super(struct super_block *sb)
854 struct ll_sb_info *sbi = ll_s2sbi(sb);
859 obd_fid_fini(sbi->ll_dt_exp->exp_obd);
860 obd_disconnect(sbi->ll_dt_exp);
861 sbi->ll_dt_exp = NULL;
863 ll_debugfs_unregister_super(sb);
865 obd_fid_fini(sbi->ll_md_exp->exp_obd);
866 obd_disconnect(sbi->ll_md_exp);
867 sbi->ll_md_exp = NULL;
872 void ll_kill_super(struct super_block *sb)
874 struct ll_sb_info *sbi;
878 if (!(sb->s_flags & SB_ACTIVE))
882 /* we need restore s_dev from changed for clustred NFS before put_super
883 * because new kernels have cached s_dev and change sb->s_dev in
884 * put_super not affected real removing devices */
886 sb->s_dev = sbi->ll_sdev_orig;
888 /* wait running statahead threads to quit */
889 while (atomic_read(&sbi->ll_sa_running) > 0)
890 schedule_timeout_uninterruptible(
891 cfs_time_seconds(1) >> 3);
897 /* Since we use this table for ll_sbi_flags_seq_show make
898 * sure what you want displayed for a specific token that
899 * is listed more than once below be listed first. For
900 * example we want "checksum" displayed, not "nochecksum"
903 static const match_table_t ll_sbi_flags_name = {
904 {LL_SBI_NOLCK, "nolock"},
905 {LL_SBI_CHECKSUM, "checksum"},
906 {LL_SBI_CHECKSUM, "nochecksum"},
907 {LL_SBI_LOCALFLOCK, "localflock"},
908 {LL_SBI_FLOCK, "flock"},
909 {LL_SBI_FLOCK, "noflock"},
910 {LL_SBI_USER_XATTR, "user_xattr"},
911 {LL_SBI_USER_XATTR, "nouser_xattr"},
912 {LL_SBI_LRU_RESIZE, "lruresize"},
913 {LL_SBI_LRU_RESIZE, "nolruresize"},
914 {LL_SBI_LAZYSTATFS, "lazystatfs"},
915 {LL_SBI_LAZYSTATFS, "nolazystatfs"},
916 {LL_SBI_32BIT_API, "32bitapi"},
917 {LL_SBI_USER_FID2PATH, "user_fid2path"},
918 {LL_SBI_USER_FID2PATH, "nouser_fid2path"},
919 {LL_SBI_VERBOSE, "verbose"},
920 {LL_SBI_VERBOSE, "noverbose"},
921 {LL_SBI_ALWAYS_PING, "always_ping"},
922 {LL_SBI_TEST_DUMMY_ENCRYPTION, "test_dummy_encryption"},
923 {LL_SBI_ENCRYPT, "encrypt"},
924 {LL_SBI_ENCRYPT, "noencrypt"},
925 {LL_SBI_FOREIGN_SYMLINK, "foreign_symlink=%s"},
926 {LL_SBI_NUM_MOUNT_OPT, NULL},
929 {LL_SBI_AGL_ENABLED, "agl"},
930 {LL_SBI_64BIT_HASH, "64bit_hash"},
931 {LL_SBI_LAYOUT_LOCK, "layout"},
932 {LL_SBI_XATTR_CACHE, "xattr_cache"},
933 {LL_SBI_NOROOTSQUASH, "norootsquash"},
934 {LL_SBI_FAST_READ, "fast_read"},
935 {LL_SBI_FILE_SECCTX, "file_secctx"},
936 {LL_SBI_TINY_WRITE, "tiny_write"},
937 {LL_SBI_FILE_HEAT, "file_heat"},
938 {LL_SBI_PARALLEL_DIO, "parallel_dio"},
941 int ll_sbi_flags_seq_show(struct seq_file *m, void *v)
943 struct super_block *sb = m->private;
946 for (i = 0; i < LL_SBI_NUM_FLAGS; i++) {
949 if (!test_bit(i, ll_s2sbi(sb)->ll_flags))
952 for (j = 0; j < ARRAY_SIZE(ll_sbi_flags_name); j++) {
953 if (ll_sbi_flags_name[j].token == i &&
954 ll_sbi_flags_name[j].pattern) {
956 ll_sbi_flags_name[j].pattern);
965 /* non-client-specific mount options are parsed in lmd_parse */
966 static int ll_options(char *options, struct super_block *sb)
968 struct ll_sb_info *sbi = ll_s2sbi(sb);
969 char *s2, *s1, *opts;
975 /* Don't stomp on lmd_opts */
976 opts = kstrdup(options, GFP_KERNEL);
982 CDEBUG(D_CONFIG, "Parsing opts %s\n", options);
984 while ((s1 = strsep(&opts, ",")) != NULL) {
985 substring_t args[MAX_OPT_ARGS];
986 bool turn_off = false;
992 CDEBUG(D_SUPER, "next opt=%s\n", s1);
994 if (strncmp(s1, "no", 2) == 0)
998 * Initialize args struct so we know whether arg was
999 * found; some options take optional arguments.
1002 args[0].from = NULL;
1003 token = match_token(s1, ll_sbi_flags_name, args);
1004 if (token == LL_SBI_NUM_MOUNT_OPT) {
1005 if (match_wildcard("context", s1) ||
1006 match_wildcard("fscontext", s1) ||
1007 match_wildcard("defcontext", s1) ||
1008 match_wildcard("rootcontext",s1))
1011 LCONSOLE_ERROR_MSG(0x152,
1012 "Unknown option '%s', won't mount.\n",
1019 case LL_SBI_32BIT_API:
1020 case LL_SBI_64BIT_HASH:
1021 case LL_SBI_ALWAYS_PING:
1022 set_bit(token, sbi->ll_flags);
1026 clear_bit(LL_SBI_LOCALFLOCK, sbi->ll_flags);
1028 clear_bit(LL_SBI_FLOCK, sbi->ll_flags);
1030 set_bit(token, sbi->ll_flags);
1033 case LL_SBI_LOCALFLOCK:
1034 clear_bit(LL_SBI_FLOCK, sbi->ll_flags);
1035 set_bit(token, sbi->ll_flags);
1038 case LL_SBI_CHECKSUM:
1039 sbi->ll_checksum_set = 1;
1041 case LL_SBI_USER_XATTR:
1042 case LL_SBI_USER_FID2PATH:
1043 case LL_SBI_LRU_RESIZE:
1044 case LL_SBI_LAZYSTATFS:
1045 case LL_SBI_VERBOSE:
1047 clear_bit(token, sbi->ll_flags);
1049 set_bit(token, sbi->ll_flags);
1051 case LL_SBI_TEST_DUMMY_ENCRYPTION: {
1052 #ifdef HAVE_LUSTRE_CRYPTO
1053 set_bit(token, sbi->ll_flags);
1055 LCONSOLE_WARN("Test dummy encryption mount option ignored: encryption not supported\n");
1059 case LL_SBI_ENCRYPT:
1060 #ifdef HAVE_LUSTRE_CRYPTO
1062 clear_bit(token, sbi->ll_flags);
1064 set_bit(token, sbi->ll_flags);
1066 LCONSOLE_WARN("noencrypt or encrypt mount option ignored: encryption not supported\n");
1069 case LL_SBI_FOREIGN_SYMLINK:
1070 /* non-default prefix provided ? */
1075 /* path must be absolute */
1076 if (args->from[0] != '/') {
1077 LCONSOLE_ERROR_MSG(0x152,
1078 "foreign prefix '%s' must be an absolute path\n",
1083 old_len = sbi->ll_foreign_symlink_prefix_size;
1084 old = sbi->ll_foreign_symlink_prefix;
1085 /* alloc for path length and '\0' */
1086 sbi->ll_foreign_symlink_prefix = match_strdup(args);
1087 if (!sbi->ll_foreign_symlink_prefix) {
1088 /* restore previous */
1089 sbi->ll_foreign_symlink_prefix = old;
1090 sbi->ll_foreign_symlink_prefix_size =
1094 sbi->ll_foreign_symlink_prefix_size =
1095 args->to - args->from + 1;
1096 OBD_ALLOC_POST(sbi->ll_foreign_symlink_prefix,
1097 sbi->ll_foreign_symlink_prefix_size,
1100 OBD_FREE(old, old_len);
1102 /* enable foreign symlink support */
1103 set_bit(token, sbi->ll_flags);
1105 LCONSOLE_ERROR_MSG(0x152,
1106 "invalid %s option\n", s1);
1117 void ll_lli_init(struct ll_inode_info *lli)
1119 lli->lli_inode_magic = LLI_INODE_MAGIC;
1121 rwlock_init(&lli->lli_lock);
1122 lli->lli_posix_acl = NULL;
1123 /* Do not set lli_fid, it has been initialized already. */
1124 fid_zero(&lli->lli_pfid);
1125 lli->lli_mds_read_och = NULL;
1126 lli->lli_mds_write_och = NULL;
1127 lli->lli_mds_exec_och = NULL;
1128 lli->lli_open_fd_read_count = 0;
1129 lli->lli_open_fd_write_count = 0;
1130 lli->lli_open_fd_exec_count = 0;
1131 mutex_init(&lli->lli_och_mutex);
1132 spin_lock_init(&lli->lli_agl_lock);
1133 spin_lock_init(&lli->lli_layout_lock);
1134 ll_layout_version_set(lli, CL_LAYOUT_GEN_NONE);
1135 lli->lli_clob = NULL;
1137 init_rwsem(&lli->lli_xattrs_list_rwsem);
1138 mutex_init(&lli->lli_xattrs_enq_lock);
1140 LASSERT(lli->lli_vfs_inode.i_mode != 0);
1141 if (S_ISDIR(lli->lli_vfs_inode.i_mode)) {
1142 lli->lli_opendir_key = NULL;
1143 lli->lli_sai = NULL;
1144 spin_lock_init(&lli->lli_sa_lock);
1145 lli->lli_opendir_pid = 0;
1146 lli->lli_sa_enabled = 0;
1147 init_rwsem(&lli->lli_lsm_sem);
1149 mutex_init(&lli->lli_size_mutex);
1150 mutex_init(&lli->lli_setattr_mutex);
1151 lli->lli_symlink_name = NULL;
1152 ll_trunc_sem_init(&lli->lli_trunc_sem);
1153 range_lock_tree_init(&lli->lli_write_tree);
1154 init_rwsem(&lli->lli_glimpse_sem);
1155 lli->lli_glimpse_time = ktime_set(0, 0);
1156 INIT_LIST_HEAD(&lli->lli_agl_list);
1157 lli->lli_agl_index = 0;
1158 lli->lli_async_rc = 0;
1159 spin_lock_init(&lli->lli_heat_lock);
1160 obd_heat_clear(lli->lli_heat_instances, OBD_HEAT_COUNT);
1161 lli->lli_heat_flags = 0;
1162 mutex_init(&lli->lli_pcc_lock);
1163 lli->lli_pcc_state = PCC_STATE_FL_NONE;
1164 lli->lli_pcc_inode = NULL;
1165 lli->lli_pcc_dsflags = PCC_DATASET_INVALID;
1166 lli->lli_pcc_generation = 0;
1167 mutex_init(&lli->lli_group_mutex);
1168 lli->lli_group_users = 0;
1169 lli->lli_group_gid = 0;
1171 mutex_init(&lli->lli_layout_mutex);
1172 memset(lli->lli_jobid, 0, sizeof(lli->lli_jobid));
1173 /* ll_cl_context initialize */
1174 INIT_LIST_HEAD(&lli->lli_lccs);
1177 #define MAX_STRING_SIZE 128
1179 #ifndef HAVE_SUPER_SETUP_BDI_NAME
1181 #define LSI_BDI_INITIALIZED 0x00400000
1183 #ifndef HAVE_BDI_CAP_MAP_COPY
1184 # define BDI_CAP_MAP_COPY 0
1187 static int super_setup_bdi_name(struct super_block *sb, char *fmt, ...)
1189 struct lustre_sb_info *lsi = s2lsi(sb);
1190 char buf[MAX_STRING_SIZE];
1194 err = bdi_init(&lsi->lsi_bdi);
1198 lsi->lsi_flags |= LSI_BDI_INITIALIZED;
1199 lsi->lsi_bdi.capabilities = BDI_CAP_MAP_COPY;
1200 lsi->lsi_bdi.name = "lustre";
1201 va_start(args, fmt);
1202 vsnprintf(buf, MAX_STRING_SIZE, fmt, args);
1204 err = bdi_register(&lsi->lsi_bdi, NULL, "%s", buf);
1207 sb->s_bdi = &lsi->lsi_bdi;
1211 #endif /* !HAVE_SUPER_SETUP_BDI_NAME */
1213 int ll_fill_super(struct super_block *sb)
1215 struct lustre_profile *lprof = NULL;
1216 struct lustre_sb_info *lsi = s2lsi(sb);
1217 struct ll_sb_info *sbi = NULL;
1218 char *dt = NULL, *md = NULL;
1219 char *profilenm = get_profile_name(sb);
1220 struct config_llog_instance *cfg;
1221 /* %p for void* in printf needs 16+2 characters: 0xffffffffffffffff */
1222 const int instlen = LUSTRE_MAXINSTANCE + 2;
1223 unsigned long cfg_instance = ll_get_cfg_instance(sb);
1224 char name[MAX_STRING_SIZE];
1233 /* for ASLR, to map between cfg_instance and hashed ptr */
1234 CDEBUG(D_VFSTRACE, "VFS Op: cfg_instance %s-%016lx (sb %p)\n",
1235 profilenm, cfg_instance, sb);
1237 OBD_RACE(OBD_FAIL_LLITE_RACE_MOUNT);
1241 GOTO(out_free_cfg, err = -ENOMEM);
1243 /* client additional sb info */
1244 lsi->lsi_llsbi = sbi = ll_init_sbi();
1246 GOTO(out_free_cfg, err = PTR_ERR(sbi));
1248 err = ll_options(lsi->lsi_lmd->lmd_opts, sb);
1250 GOTO(out_free_cfg, err);
1252 /* kernel >= 2.6.38 store dentry operations in sb->s_d_op. */
1253 sb->s_d_op = &ll_d_ops;
1256 generate_random_uuid(uuid.b);
1257 snprintf(sbi->ll_sb_uuid.uuid, sizeof(sbi->ll_sb_uuid), "%pU", uuid.b);
1259 CDEBUG(D_CONFIG, "llite sb uuid: %s\n", sbi->ll_sb_uuid.uuid);
1262 len = strlen(profilenm);
1263 ptr = strrchr(profilenm, '-');
1264 if (ptr && (strcmp(ptr, "-client") == 0))
1267 if (len > LUSTRE_MAXFSNAME) {
1268 if (unlikely(len >= MAX_STRING_SIZE))
1269 len = MAX_STRING_SIZE - 1;
1270 strncpy(name, profilenm, len);
1272 err = -ENAMETOOLONG;
1273 CERROR("%s: fsname longer than %u characters: rc = %d\n",
1274 name, LUSTRE_MAXFSNAME, err);
1275 GOTO(out_free_cfg, err);
1277 strncpy(sbi->ll_fsname, profilenm, len);
1278 sbi->ll_fsname[len] = '\0';
1281 snprintf(name, sizeof(name), "%.*s-%016lx", len,
1282 profilenm, cfg_instance);
1284 err = super_setup_bdi_name(sb, "%s", name);
1286 GOTO(out_free_cfg, err);
1288 /* Call ll_debugfs_register_super() before lustre_process_log()
1289 * so that "llite.*.*" params can be processed correctly.
1291 err = ll_debugfs_register_super(sb, name);
1293 CERROR("%s: could not register mountpoint in llite: rc = %d\n",
1294 sbi->ll_fsname, err);
1298 /* The cfg_instance is a value unique to this super, in case some
1299 * joker tries to mount the same fs at two mount points.
1301 cfg->cfg_instance = cfg_instance;
1302 cfg->cfg_uuid = lsi->lsi_llsbi->ll_sb_uuid;
1303 cfg->cfg_callback = class_config_llog_handler;
1304 cfg->cfg_sub_clds = CONFIG_SUB_CLIENT;
1305 /* set up client obds */
1306 err = lustre_process_log(sb, profilenm, cfg);
1308 GOTO(out_debugfs, err);
1310 /* Profile set with LCFG_MOUNTOPT so we can find our mdc and osc obds */
1311 lprof = class_get_profile(profilenm);
1312 if (lprof == NULL) {
1313 LCONSOLE_ERROR_MSG(0x156, "The client profile '%s' could not be"
1314 " read from the MGS. Does that filesystem "
1315 "exist?\n", profilenm);
1316 GOTO(out_debugfs, err = -EINVAL);
1318 CDEBUG(D_CONFIG, "Found profile %s: mdc=%s osc=%s\n", profilenm,
1319 lprof->lp_md, lprof->lp_dt);
1321 dt_len = strlen(lprof->lp_dt) + instlen + 2;
1322 OBD_ALLOC(dt, dt_len);
1324 GOTO(out_profile, err = -ENOMEM);
1325 snprintf(dt, dt_len - 1, "%s-%016lx", lprof->lp_dt, cfg_instance);
1327 md_len = strlen(lprof->lp_md) + instlen + 2;
1328 OBD_ALLOC(md, md_len);
1330 GOTO(out_free_dt, err = -ENOMEM);
1331 snprintf(md, md_len - 1, "%s-%016lx", lprof->lp_md, cfg_instance);
1333 /* connections, registrations, sb setup */
1334 err = client_common_fill_super(sb, md, dt);
1336 GOTO(out_free_md, err);
1338 sbi->ll_client_common_fill_super_succeeded = 1;
1342 OBD_FREE(md, md_len);
1345 OBD_FREE(dt, dt_len);
1348 class_put_profile(lprof);
1351 ll_debugfs_unregister_super(sb);
1358 else if (test_bit(LL_SBI_VERBOSE, sbi->ll_flags))
1359 LCONSOLE_WARN("Mounted %s\n", profilenm);
1361 } /* ll_fill_super */
1363 void ll_put_super(struct super_block *sb)
1365 struct config_llog_instance cfg, params_cfg;
1366 struct obd_device *obd;
1367 struct lustre_sb_info *lsi = s2lsi(sb);
1368 struct ll_sb_info *sbi = ll_s2sbi(sb);
1369 char *profilenm = get_profile_name(sb);
1370 unsigned long cfg_instance = ll_get_cfg_instance(sb);
1372 int next, force = 1, rc = 0;
1376 GOTO(out_no_sbi, 0);
1378 /* Should replace instance_id with something better for ASLR */
1379 CDEBUG(D_VFSTRACE, "VFS Op: cfg_instance %s-%016lx (sb %p)\n",
1380 profilenm, cfg_instance, sb);
1382 cfg.cfg_instance = cfg_instance;
1383 lustre_end_log(sb, profilenm, &cfg);
1385 params_cfg.cfg_instance = cfg_instance;
1386 lustre_end_log(sb, PARAMS_FILENAME, ¶ms_cfg);
1388 if (sbi->ll_md_exp) {
1389 obd = class_exp2obd(sbi->ll_md_exp);
1391 force = obd->obd_force;
1394 /* Wait for unstable pages to be committed to stable storage */
1396 rc = l_wait_event_abortable(
1397 sbi->ll_cache->ccc_unstable_waitq,
1398 atomic_long_read(&sbi->ll_cache->ccc_unstable_nr) == 0);
1401 ccc_count = atomic_long_read(&sbi->ll_cache->ccc_unstable_nr);
1402 if (force == 0 && rc != -ERESTARTSYS)
1403 LASSERTF(ccc_count == 0, "count: %li\n", ccc_count);
1405 /* We need to set force before the lov_disconnect in
1406 * lustre_common_put_super, since l_d cleans up osc's as well.
1410 while ((obd = class_devices_in_group(&sbi->ll_sb_uuid,
1412 obd->obd_force = force;
1416 if (sbi->ll_client_common_fill_super_succeeded) {
1417 /* Only if client_common_fill_super succeeded */
1418 client_common_put_super(sb);
1422 while ((obd = class_devices_in_group(&sbi->ll_sb_uuid, &next)))
1423 class_manual_cleanup(obd);
1425 if (test_bit(LL_SBI_VERBOSE, sbi->ll_flags))
1426 LCONSOLE_WARN("Unmounted %s\n", profilenm ? profilenm : "");
1429 class_del_profile(profilenm);
1431 #ifndef HAVE_SUPER_SETUP_BDI_NAME
1432 if (lsi->lsi_flags & LSI_BDI_INITIALIZED) {
1433 bdi_destroy(&lsi->lsi_bdi);
1434 lsi->lsi_flags &= ~LSI_BDI_INITIALIZED;
1439 lsi->lsi_llsbi = NULL;
1441 lustre_common_put_super(sb);
1443 cl_env_cache_purge(~0);
1446 } /* client_put_super */
1448 struct inode *ll_inode_from_resource_lock(struct ldlm_lock *lock)
1450 struct inode *inode = NULL;
1452 /* NOTE: we depend on atomic igrab() -bzzz */
1453 lock_res_and_lock(lock);
1454 if (lock->l_resource->lr_lvb_inode) {
1455 struct ll_inode_info * lli;
1456 lli = ll_i2info(lock->l_resource->lr_lvb_inode);
1457 if (lli->lli_inode_magic == LLI_INODE_MAGIC) {
1458 inode = igrab(lock->l_resource->lr_lvb_inode);
1460 inode = lock->l_resource->lr_lvb_inode;
1461 LDLM_DEBUG_LIMIT(inode->i_state & I_FREEING ? D_INFO :
1462 D_WARNING, lock, "lr_lvb_inode %p is "
1463 "bogus: magic %08x",
1464 lock->l_resource->lr_lvb_inode,
1465 lli->lli_inode_magic);
1469 unlock_res_and_lock(lock);
1473 void ll_dir_clear_lsm_md(struct inode *inode)
1475 struct ll_inode_info *lli = ll_i2info(inode);
1477 LASSERT(S_ISDIR(inode->i_mode));
1479 if (lli->lli_lsm_md) {
1480 lmv_free_memmd(lli->lli_lsm_md);
1481 lli->lli_lsm_md = NULL;
1484 if (lli->lli_default_lsm_md) {
1485 lmv_free_memmd(lli->lli_default_lsm_md);
1486 lli->lli_default_lsm_md = NULL;
1490 static struct inode *ll_iget_anon_dir(struct super_block *sb,
1491 const struct lu_fid *fid,
1492 struct lustre_md *md)
1494 struct ll_sb_info *sbi = ll_s2sbi(sb);
1495 struct ll_inode_info *lli;
1496 struct mdt_body *body = md->body;
1497 struct inode *inode;
1503 ino = cl_fid_build_ino(fid, test_bit(LL_SBI_32BIT_API, sbi->ll_flags));
1504 inode = iget_locked(sb, ino);
1505 if (inode == NULL) {
1506 CERROR("%s: failed get simple inode "DFID": rc = -ENOENT\n",
1507 sbi->ll_fsname, PFID(fid));
1508 RETURN(ERR_PTR(-ENOENT));
1511 lli = ll_i2info(inode);
1512 if (inode->i_state & I_NEW) {
1513 inode->i_mode = (inode->i_mode & ~S_IFMT) |
1514 (body->mbo_mode & S_IFMT);
1515 LASSERTF(S_ISDIR(inode->i_mode), "Not slave inode "DFID"\n",
1518 inode->i_mtime.tv_sec = 0;
1519 inode->i_atime.tv_sec = 0;
1520 inode->i_ctime.tv_sec = 0;
1523 #ifdef HAVE_BACKING_DEV_INFO
1524 /* initializing backing dev info. */
1525 inode->i_mapping->backing_dev_info =
1526 &s2lsi(inode->i_sb)->lsi_bdi;
1528 inode->i_op = &ll_dir_inode_operations;
1529 inode->i_fop = &ll_dir_operations;
1530 lli->lli_fid = *fid;
1533 /* master object FID */
1534 lli->lli_pfid = body->mbo_fid1;
1535 CDEBUG(D_INODE, "lli %p slave "DFID" master "DFID"\n",
1536 lli, PFID(fid), PFID(&lli->lli_pfid));
1537 unlock_new_inode(inode);
1539 /* in directory restripe/auto-split, a directory will be
1540 * transformed to a stripe if it's plain, set its pfid here,
1541 * otherwise ll_lock_cancel_bits() can't find the master inode.
1543 lli->lli_pfid = body->mbo_fid1;
1549 static int ll_init_lsm_md(struct inode *inode, struct lustre_md *md)
1552 struct lmv_stripe_md *lsm = md->lmv;
1553 struct ll_inode_info *lli = ll_i2info(inode);
1556 LASSERT(lsm != NULL);
1558 CDEBUG(D_INODE, "%s: "DFID" set dir layout:\n",
1559 ll_i2sbi(inode)->ll_fsname, PFID(&lli->lli_fid));
1560 lsm_md_dump(D_INODE, lsm);
1562 if (!lmv_dir_striped(lsm))
1565 /* XXX sigh, this lsm_root initialization should be in
1566 * LMV layer, but it needs ll_iget right now, so we
1567 * put this here right now. */
1568 for (i = 0; i < lsm->lsm_md_stripe_count; i++) {
1569 fid = &lsm->lsm_md_oinfo[i].lmo_fid;
1570 LASSERT(lsm->lsm_md_oinfo[i].lmo_root == NULL);
1572 if (!fid_is_sane(fid))
1575 /* Unfortunately ll_iget will call ll_update_inode,
1576 * where the initialization of slave inode is slightly
1577 * different, so it reset lsm_md to NULL to avoid
1578 * initializing lsm for slave inode. */
1579 lsm->lsm_md_oinfo[i].lmo_root =
1580 ll_iget_anon_dir(inode->i_sb, fid, md);
1581 if (IS_ERR(lsm->lsm_md_oinfo[i].lmo_root)) {
1582 int rc = PTR_ERR(lsm->lsm_md_oinfo[i].lmo_root);
1584 lsm->lsm_md_oinfo[i].lmo_root = NULL;
1586 iput(lsm->lsm_md_oinfo[i].lmo_root);
1587 lsm->lsm_md_oinfo[i].lmo_root = NULL;
1593 lli->lli_lsm_md = lsm;
1598 static void ll_update_default_lsm_md(struct inode *inode, struct lustre_md *md)
1600 struct ll_inode_info *lli = ll_i2info(inode);
1604 if (!md->default_lmv) {
1605 /* clear default lsm */
1606 if (lli->lli_default_lsm_md) {
1607 down_write(&lli->lli_lsm_sem);
1608 if (lli->lli_default_lsm_md) {
1609 lmv_free_memmd(lli->lli_default_lsm_md);
1610 lli->lli_default_lsm_md = NULL;
1612 up_write(&lli->lli_lsm_sem);
1617 if (lli->lli_default_lsm_md) {
1618 /* do nonthing if default lsm isn't changed */
1619 down_read(&lli->lli_lsm_sem);
1620 if (lli->lli_default_lsm_md &&
1621 lsm_md_eq(lli->lli_default_lsm_md, md->default_lmv)) {
1622 up_read(&lli->lli_lsm_sem);
1625 up_read(&lli->lli_lsm_sem);
1628 down_write(&lli->lli_lsm_sem);
1629 if (lli->lli_default_lsm_md)
1630 lmv_free_memmd(lli->lli_default_lsm_md);
1631 lli->lli_default_lsm_md = md->default_lmv;
1632 lsm_md_dump(D_INODE, md->default_lmv);
1633 md->default_lmv = NULL;
1634 up_write(&lli->lli_lsm_sem);
1638 static int ll_update_lsm_md(struct inode *inode, struct lustre_md *md)
1640 struct ll_inode_info *lli = ll_i2info(inode);
1641 struct lmv_stripe_md *lsm = md->lmv;
1642 struct cl_attr *attr;
1647 LASSERT(S_ISDIR(inode->i_mode));
1648 CDEBUG(D_INODE, "update lsm %p of "DFID"\n", lli->lli_lsm_md,
1649 PFID(ll_inode2fid(inode)));
1651 /* update default LMV */
1652 if (md->default_lmv)
1653 ll_update_default_lsm_md(inode, md);
1655 /* after dir migration/restripe, a stripe may be turned into a
1656 * directory, in this case, zero out its lli_pfid.
1658 if (unlikely(fid_is_norm(&lli->lli_pfid)))
1659 fid_zero(&lli->lli_pfid);
1662 * no striped information from request, lustre_md from req does not
1663 * include stripeEA, see ll_md_setattr()
1669 * normally dir layout doesn't change, only take read lock to check
1670 * that to avoid blocking other MD operations.
1672 down_read(&lli->lli_lsm_sem);
1674 /* some current lookup initialized lsm, and unchanged */
1675 if (lli->lli_lsm_md && lsm_md_eq(lli->lli_lsm_md, lsm))
1676 GOTO(unlock, rc = 0);
1678 /* if dir layout doesn't match, check whether version is increased,
1679 * which means layout is changed, this happens in dir split/merge and
1682 * foreign LMV should not change.
1684 if (lli->lli_lsm_md && lmv_dir_striped(lli->lli_lsm_md) &&
1685 lsm->lsm_md_layout_version <=
1686 lli->lli_lsm_md->lsm_md_layout_version) {
1687 CERROR("%s: "DFID" dir layout mismatch:\n",
1688 ll_i2sbi(inode)->ll_fsname, PFID(&lli->lli_fid));
1689 lsm_md_dump(D_ERROR, lli->lli_lsm_md);
1690 lsm_md_dump(D_ERROR, lsm);
1691 GOTO(unlock, rc = -EINVAL);
1694 up_read(&lli->lli_lsm_sem);
1695 down_write(&lli->lli_lsm_sem);
1696 /* clear existing lsm */
1697 if (lli->lli_lsm_md) {
1698 lmv_free_memmd(lli->lli_lsm_md);
1699 lli->lli_lsm_md = NULL;
1702 rc = ll_init_lsm_md(inode, md);
1703 up_write(&lli->lli_lsm_sem);
1708 /* set md->lmv to NULL, so the following free lustre_md will not free
1713 /* md_merge_attr() may take long, since lsm is already set, switch to
1716 down_read(&lli->lli_lsm_sem);
1718 if (!lmv_dir_striped(lli->lli_lsm_md))
1719 GOTO(unlock, rc = 0);
1721 OBD_ALLOC_PTR(attr);
1723 GOTO(unlock, rc = -ENOMEM);
1725 /* validate the lsm */
1726 rc = md_merge_attr(ll_i2mdexp(inode), lli->lli_lsm_md, attr,
1727 ll_md_blocking_ast);
1729 if (md->body->mbo_valid & OBD_MD_FLNLINK)
1730 md->body->mbo_nlink = attr->cat_nlink;
1731 if (md->body->mbo_valid & OBD_MD_FLSIZE)
1732 md->body->mbo_size = attr->cat_size;
1733 if (md->body->mbo_valid & OBD_MD_FLATIME)
1734 md->body->mbo_atime = attr->cat_atime;
1735 if (md->body->mbo_valid & OBD_MD_FLCTIME)
1736 md->body->mbo_ctime = attr->cat_ctime;
1737 if (md->body->mbo_valid & OBD_MD_FLMTIME)
1738 md->body->mbo_mtime = attr->cat_mtime;
1744 up_read(&lli->lli_lsm_sem);
1749 void ll_clear_inode(struct inode *inode)
1751 struct ll_inode_info *lli = ll_i2info(inode);
1752 struct ll_sb_info *sbi = ll_i2sbi(inode);
1756 CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p)\n",
1757 PFID(ll_inode2fid(inode)), inode);
1759 if (S_ISDIR(inode->i_mode)) {
1760 /* these should have been cleared in ll_file_release */
1761 LASSERT(lli->lli_opendir_key == NULL);
1762 LASSERT(lli->lli_sai == NULL);
1763 LASSERT(lli->lli_opendir_pid == 0);
1765 pcc_inode_free(inode);
1768 md_null_inode(sbi->ll_md_exp, ll_inode2fid(inode));
1770 LASSERT(!lli->lli_open_fd_write_count);
1771 LASSERT(!lli->lli_open_fd_read_count);
1772 LASSERT(!lli->lli_open_fd_exec_count);
1774 if (lli->lli_mds_write_och)
1775 ll_md_real_close(inode, FMODE_WRITE);
1776 if (lli->lli_mds_exec_och)
1777 ll_md_real_close(inode, FMODE_EXEC);
1778 if (lli->lli_mds_read_och)
1779 ll_md_real_close(inode, FMODE_READ);
1781 if (S_ISLNK(inode->i_mode) && lli->lli_symlink_name) {
1782 OBD_FREE(lli->lli_symlink_name,
1783 strlen(lli->lli_symlink_name) + 1);
1784 lli->lli_symlink_name = NULL;
1787 ll_xattr_cache_destroy(inode);
1789 forget_all_cached_acls(inode);
1791 lli->lli_inode_magic = LLI_INODE_DEAD;
1793 if (S_ISDIR(inode->i_mode))
1794 ll_dir_clear_lsm_md(inode);
1795 else if (S_ISREG(inode->i_mode) && !is_bad_inode(inode))
1796 LASSERT(list_empty(&lli->lli_agl_list));
1799 * XXX This has to be done before lsm is freed below, because
1800 * cl_object still uses inode lsm.
1802 cl_inode_fini(inode);
1804 llcrypt_put_encryption_info(inode);
1809 static int ll_md_setattr(struct dentry *dentry, struct md_op_data *op_data)
1811 struct lustre_md md;
1812 struct inode *inode = dentry->d_inode;
1813 struct ll_sb_info *sbi = ll_i2sbi(inode);
1814 struct ptlrpc_request *request = NULL;
1819 op_data = ll_prep_md_op_data(op_data, inode, NULL, NULL, 0, 0,
1820 LUSTRE_OPC_ANY, NULL);
1821 if (IS_ERR(op_data))
1822 RETURN(PTR_ERR(op_data));
1824 /* If this is a chgrp of a regular file, we want to reserve enough
1825 * quota to cover the entire file size.
1827 if (S_ISREG(inode->i_mode) && op_data->op_attr.ia_valid & ATTR_GID &&
1828 from_kgid(&init_user_ns, op_data->op_attr.ia_gid) !=
1829 from_kgid(&init_user_ns, inode->i_gid)) {
1830 op_data->op_xvalid |= OP_XVALID_BLOCKS;
1831 op_data->op_attr_blocks = inode->i_blocks;
1835 rc = md_setattr(sbi->ll_md_exp, op_data, NULL, 0, &request);
1837 ptlrpc_req_finished(request);
1838 if (rc == -ENOENT) {
1840 /* Unlinked special device node? Or just a race?
1841 * Pretend we done everything. */
1842 if (!S_ISREG(inode->i_mode) &&
1843 !S_ISDIR(inode->i_mode)) {
1844 ia_valid = op_data->op_attr.ia_valid;
1845 op_data->op_attr.ia_valid &= ~TIMES_SET_FLAGS;
1846 rc = simple_setattr(dentry, &op_data->op_attr);
1847 op_data->op_attr.ia_valid = ia_valid;
1849 } else if (rc != -EPERM && rc != -EACCES && rc != -ETXTBSY) {
1850 CERROR("md_setattr fails: rc = %d\n", rc);
1855 rc = md_get_lustre_md(sbi->ll_md_exp, &request->rq_pill, sbi->ll_dt_exp,
1856 sbi->ll_md_exp, &md);
1858 ptlrpc_req_finished(request);
1862 ia_valid = op_data->op_attr.ia_valid;
1863 /* inode size will be in ll_setattr_ost, can't do it now since dirty
1864 * cache is not cleared yet. */
1865 op_data->op_attr.ia_valid &= ~(TIMES_SET_FLAGS | ATTR_SIZE);
1866 if (S_ISREG(inode->i_mode))
1868 rc = simple_setattr(dentry, &op_data->op_attr);
1869 if (S_ISREG(inode->i_mode))
1870 inode_unlock(inode);
1871 op_data->op_attr.ia_valid = ia_valid;
1873 rc = ll_update_inode(inode, &md);
1874 ptlrpc_req_finished(request);
1880 * Zero portion of page that is part of @inode.
1881 * This implies, if necessary:
1882 * - taking cl_lock on range corresponding to concerned page
1883 * - grabbing vm page
1884 * - associating cl_page
1885 * - proceeding to clio read
1886 * - zeroing range in page
1887 * - proceeding to cl_page flush
1888 * - releasing cl_lock
1890 * \param[in] inode inode
1891 * \param[in] index page index
1892 * \param[in] offset offset in page to start zero from
1893 * \param[in] len len to zero
1895 * \retval 0 on success
1896 * \retval negative errno on failure
1898 int ll_io_zero_page(struct inode *inode, pgoff_t index, pgoff_t offset,
1901 struct ll_inode_info *lli = ll_i2info(inode);
1902 struct cl_object *clob = lli->lli_clob;
1904 struct lu_env *env = NULL;
1905 struct cl_io *io = NULL;
1906 struct cl_page *clpage = NULL;
1907 struct page *vmpage = NULL;
1908 unsigned from = index << PAGE_SHIFT;
1909 struct cl_lock *lock = NULL;
1910 struct cl_lock_descr *descr = NULL;
1911 struct cl_2queue *queue = NULL;
1912 struct cl_sync_io *anchor = NULL;
1913 bool holdinglock = false;
1914 bool lockedbymyself = true;
1919 env = cl_env_get(&refcheck);
1921 RETURN(PTR_ERR(env));
1923 io = vvp_env_thread_io(env);
1925 rc = cl_io_rw_init(env, io, CIT_WRITE, from, PAGE_SIZE);
1929 lock = vvp_env_lock(env);
1930 descr = &lock->cll_descr;
1931 descr->cld_obj = io->ci_obj;
1932 descr->cld_start = cl_index(io->ci_obj, from);
1933 descr->cld_end = cl_index(io->ci_obj, from + PAGE_SIZE - 1);
1934 descr->cld_mode = CLM_WRITE;
1935 descr->cld_enq_flags = CEF_MUST | CEF_NONBLOCK;
1937 /* request lock for page */
1938 rc = cl_lock_request(env, io, lock);
1939 /* -ECANCELED indicates a matching lock with a different extent
1940 * was already present, and -EEXIST indicates a matching lock
1941 * on exactly the same extent was already present.
1942 * In both cases it means we are covered.
1944 if (rc == -ECANCELED || rc == -EEXIST)
1952 vmpage = grab_cache_page_nowait(inode->i_mapping, index);
1954 GOTO(rellock, rc = -EOPNOTSUPP);
1956 if (!PageDirty(vmpage)) {
1957 /* associate cl_page */
1958 clpage = cl_page_find(env, clob, vmpage->index,
1959 vmpage, CPT_CACHEABLE);
1961 GOTO(pagefini, rc = PTR_ERR(clpage));
1963 cl_page_assume(env, io, clpage);
1966 if (!PageUptodate(vmpage) && !PageDirty(vmpage) &&
1967 !PageWriteback(vmpage)) {
1969 /* set PagePrivate2 to detect special case of empty page
1970 * in osc_brw_fini_request()
1972 SetPagePrivate2(vmpage);
1973 rc = ll_io_read_page(env, io, clpage, NULL);
1974 if (!PagePrivate2(vmpage))
1975 /* PagePrivate2 was cleared in osc_brw_fini_request()
1976 * meaning we read an empty page. In this case, in order
1977 * to avoid allocating unnecessary block in truncated
1978 * file, we must not zero and write as below. Subsequent
1979 * server-side truncate will handle things correctly.
1981 GOTO(clpfini, rc = 0);
1982 ClearPagePrivate2(vmpage);
1985 lockedbymyself = trylock_page(vmpage);
1986 cl_page_assume(env, io, clpage);
1989 /* zero range in page */
1990 zero_user(vmpage, offset, len);
1992 if (holdinglock && clpage) {
1993 /* explicitly write newly modified page */
1994 queue = &io->ci_queue;
1995 cl_2queue_init(queue);
1996 anchor = &vvp_env_info(env)->vti_anchor;
1997 cl_sync_io_init(anchor, 1);
1998 clpage->cp_sync_io = anchor;
1999 cl_2queue_add(queue, clpage, true);
2000 rc = cl_io_submit_rw(env, io, CRT_WRITE, queue);
2002 GOTO(queuefini1, rc);
2003 rc = cl_sync_io_wait(env, anchor, 0);
2005 GOTO(queuefini2, rc);
2006 cl_page_assume(env, io, clpage);
2009 cl_2queue_discard(env, io, queue);
2011 cl_2queue_disown(env, io, queue);
2012 cl_2queue_fini(env, queue);
2017 cl_page_put(env, clpage);
2019 if (lockedbymyself) {
2020 unlock_page(vmpage);
2025 cl_lock_release(env, lock);
2027 cl_io_fini(env, io);
2030 cl_env_put(env, &refcheck);
2035 /* If this inode has objects allocated to it (lsm != NULL), then the OST
2036 * object(s) determine the file size and mtime. Otherwise, the MDS will
2037 * keep these values until such a time that objects are allocated for it.
2038 * We do the MDS operations first, as it is checking permissions for us.
2039 * We don't to the MDS RPC if there is nothing that we want to store there,
2040 * otherwise there is no harm in updating mtime/atime on the MDS if we are
2041 * going to do an RPC anyways.
2043 * If we are doing a truncate, we will send the mtime and ctime updates
2044 * to the OST with the punch RPC, otherwise we do an explicit setattr RPC.
2045 * I don't believe it is possible to get e.g. ATTR_MTIME_SET and ATTR_SIZE
2048 * In case of HSMimport, we only set attr on MDS.
2050 int ll_setattr_raw(struct dentry *dentry, struct iattr *attr,
2051 enum op_xvalid xvalid, bool hsm_import)
2053 struct inode *inode = dentry->d_inode;
2054 struct ll_inode_info *lli = ll_i2info(inode);
2055 struct md_op_data *op_data = NULL;
2056 ktime_t kstart = ktime_get();
2061 CDEBUG(D_VFSTRACE, "%s: setattr inode "DFID"(%p) from %llu to %llu, "
2062 "valid %x, hsm_import %d\n",
2063 ll_i2sbi(inode)->ll_fsname, PFID(&lli->lli_fid),
2064 inode, i_size_read(inode), attr->ia_size, attr->ia_valid,
2067 if (attr->ia_valid & ATTR_SIZE) {
2068 /* Check new size against VFS/VM file size limit and rlimit */
2069 rc = inode_newsize_ok(inode, attr->ia_size);
2073 /* The maximum Lustre file size is variable, based on the
2074 * OST maximum object size and number of stripes. This
2075 * needs another check in addition to the VFS check above. */
2076 if (attr->ia_size > ll_file_maxbytes(inode)) {
2077 CDEBUG(D_INODE,"file "DFID" too large %llu > %llu\n",
2078 PFID(&lli->lli_fid), attr->ia_size,
2079 ll_file_maxbytes(inode));
2083 attr->ia_valid |= ATTR_MTIME | ATTR_CTIME;
2086 /* POSIX: check before ATTR_*TIME_SET set (from inode_change_ok) */
2087 if (attr->ia_valid & TIMES_SET_FLAGS) {
2088 if ((!uid_eq(current_fsuid(), inode->i_uid)) &&
2089 !capable(CAP_FOWNER))
2093 /* We mark all of the fields "set" so MDS/OST does not re-set them */
2094 if (!(xvalid & OP_XVALID_CTIME_SET) &&
2095 (attr->ia_valid & ATTR_CTIME)) {
2096 attr->ia_ctime = current_time(inode);
2097 xvalid |= OP_XVALID_CTIME_SET;
2099 if (!(attr->ia_valid & ATTR_ATIME_SET) &&
2100 (attr->ia_valid & ATTR_ATIME)) {
2101 attr->ia_atime = current_time(inode);
2102 attr->ia_valid |= ATTR_ATIME_SET;
2104 if (!(attr->ia_valid & ATTR_MTIME_SET) &&
2105 (attr->ia_valid & ATTR_MTIME)) {
2106 attr->ia_mtime = current_time(inode);
2107 attr->ia_valid |= ATTR_MTIME_SET;
2110 if (attr->ia_valid & (ATTR_MTIME | ATTR_CTIME))
2111 CDEBUG(D_INODE, "setting mtime %lld, ctime %lld, now = %lld\n",
2112 (s64)attr->ia_mtime.tv_sec, (s64)attr->ia_ctime.tv_sec,
2113 ktime_get_real_seconds());
2115 if (S_ISREG(inode->i_mode))
2116 inode_unlock(inode);
2118 /* We always do an MDS RPC, even if we're only changing the size;
2119 * only the MDS knows whether truncate() should fail with -ETXTBUSY */
2121 OBD_ALLOC_PTR(op_data);
2122 if (op_data == NULL)
2123 GOTO(out, rc = -ENOMEM);
2125 if (!hsm_import && attr->ia_valid & ATTR_SIZE) {
2126 /* If we are changing file size, file content is
2127 * modified, flag it.
2129 xvalid |= OP_XVALID_OWNEROVERRIDE;
2130 op_data->op_bias |= MDS_DATA_MODIFIED;
2131 clear_bit(LLIF_DATA_MODIFIED, &lli->lli_flags);
2134 if (attr->ia_valid & ATTR_FILE) {
2135 struct ll_file_data *fd = attr->ia_file->private_data;
2137 if (fd->fd_lease_och)
2138 op_data->op_bias |= MDS_TRUNC_KEEP_LEASE;
2141 op_data->op_attr = *attr;
2142 op_data->op_xvalid = xvalid;
2144 rc = ll_md_setattr(dentry, op_data);
2148 if (!S_ISREG(inode->i_mode) || hsm_import)
2151 if (attr->ia_valid & (ATTR_SIZE | ATTR_ATIME | ATTR_ATIME_SET |
2152 ATTR_MTIME | ATTR_MTIME_SET | ATTR_CTIME) ||
2153 xvalid & OP_XVALID_CTIME_SET) {
2154 bool cached = false;
2156 rc = pcc_inode_setattr(inode, attr, &cached);
2159 CERROR("%s: PCC inode "DFID" setattr failed: "
2161 ll_i2sbi(inode)->ll_fsname,
2162 PFID(&lli->lli_fid), rc);
2166 unsigned int flags = 0;
2168 /* For truncate and utimes sending attributes to OSTs,
2169 * setting mtime/atime to the past will be performed
2170 * under PW [0:EOF] extent lock (new_size:EOF for
2171 * truncate). It may seem excessive to send mtime/atime
2172 * updates to OSTs when not setting times to past, but
2173 * it is necessary due to possible time
2174 * de-synchronization between MDT inode and OST objects
2176 if (S_ISREG(inode->i_mode) && IS_ENCRYPTED(inode)) {
2177 xvalid |= OP_XVALID_FLAGS;
2178 flags = LUSTRE_ENCRYPT_FL;
2179 /* Call to ll_io_zero_page is not necessary if
2180 * truncating on PAGE_SIZE boundary, because
2181 * whole pages will be wiped.
2182 * In case of Direct IO, all we need is to set
2185 if (attr->ia_valid & ATTR_SIZE &&
2186 attr->ia_size & ~PAGE_MASK &&
2187 !(attr->ia_valid & ATTR_FILE &&
2188 attr->ia_file->f_flags & O_DIRECT)) {
2190 attr->ia_size & (PAGE_SIZE - 1);
2192 rc = ll_io_zero_page(inode,
2193 attr->ia_size >> PAGE_SHIFT,
2194 offset, PAGE_SIZE - offset);
2199 rc = cl_setattr_ost(lli->lli_clob, attr, xvalid, flags);
2203 /* If the file was restored, it needs to set dirty flag.
2205 * We've already sent MDS_DATA_MODIFIED flag in
2206 * ll_md_setattr() for truncate. However, the MDT refuses to
2207 * set the HS_DIRTY flag on released files, so we have to set
2208 * it again if the file has been restored. Please check how
2209 * LLIF_DATA_MODIFIED is set in vvp_io_setattr_fini().
2211 * Please notice that if the file is not released, the previous
2212 * MDS_DATA_MODIFIED has taken effect and usually
2213 * LLIF_DATA_MODIFIED is not set(see vvp_io_setattr_fini()).
2214 * This way we can save an RPC for common open + trunc
2216 if (test_and_clear_bit(LLIF_DATA_MODIFIED, &lli->lli_flags)) {
2217 struct hsm_state_set hss = {
2218 .hss_valid = HSS_SETMASK,
2219 .hss_setmask = HS_DIRTY,
2223 rc2 = ll_hsm_state_set(inode, &hss);
2224 /* truncate and write can happen at the same time, so that
2225 * the file can be set modified even though the file is not
2226 * restored from released state, and ll_hsm_state_set() is
2227 * not applicable for the file, and rc2 < 0 is normal in this
2230 CDEBUG(D_INFO, DFID "HSM set dirty failed: rc2 = %d\n",
2231 PFID(ll_inode2fid(inode)), rc2);
2236 if (op_data != NULL)
2237 ll_finish_md_op_data(op_data);
2239 if (S_ISREG(inode->i_mode)) {
2241 if ((attr->ia_valid & ATTR_SIZE) && !hsm_import)
2242 inode_dio_wait(inode);
2243 /* Once we've got the i_mutex, it's safe to set the S_NOSEC
2244 * flag. ll_update_inode (called from ll_md_setattr), clears
2245 * inode flags, so there is a gap where S_NOSEC is not set.
2246 * This can cause a writer to take the i_mutex unnecessarily,
2247 * but this is safe to do and should be rare. */
2248 inode_has_no_xattr(inode);
2252 ll_stats_ops_tally(ll_i2sbi(inode), attr->ia_valid & ATTR_SIZE ?
2253 LPROC_LL_TRUNC : LPROC_LL_SETATTR,
2254 ktime_us_delta(ktime_get(), kstart));
2259 int ll_setattr(struct dentry *de, struct iattr *attr)
2261 int mode = de->d_inode->i_mode;
2262 enum op_xvalid xvalid = 0;
2265 rc = llcrypt_prepare_setattr(de, attr);
2269 if ((attr->ia_valid & (ATTR_CTIME|ATTR_SIZE|ATTR_MODE)) ==
2270 (ATTR_CTIME|ATTR_SIZE|ATTR_MODE))
2271 xvalid |= OP_XVALID_OWNEROVERRIDE;
2273 if (((attr->ia_valid & (ATTR_MODE|ATTR_FORCE|ATTR_SIZE)) ==
2274 (ATTR_SIZE|ATTR_MODE)) &&
2275 (((mode & S_ISUID) && !(attr->ia_mode & S_ISUID)) ||
2276 (((mode & (S_ISGID|S_IXGRP)) == (S_ISGID|S_IXGRP)) &&
2277 !(attr->ia_mode & S_ISGID))))
2278 attr->ia_valid |= ATTR_FORCE;
2280 if ((attr->ia_valid & ATTR_MODE) &&
2282 !(attr->ia_mode & S_ISUID) &&
2283 !(attr->ia_valid & ATTR_KILL_SUID))
2284 attr->ia_valid |= ATTR_KILL_SUID;
2286 if ((attr->ia_valid & ATTR_MODE) &&
2287 ((mode & (S_ISGID|S_IXGRP)) == (S_ISGID|S_IXGRP)) &&
2288 !(attr->ia_mode & S_ISGID) &&
2289 !(attr->ia_valid & ATTR_KILL_SGID))
2290 attr->ia_valid |= ATTR_KILL_SGID;
2292 return ll_setattr_raw(de, attr, xvalid, false);
2295 int ll_statfs_internal(struct ll_sb_info *sbi, struct obd_statfs *osfs,
2298 struct obd_statfs obd_osfs = { 0 };
2303 max_age = ktime_get_seconds() - sbi->ll_statfs_max_age;
2305 if (test_bit(LL_SBI_LAZYSTATFS, sbi->ll_flags))
2306 flags |= OBD_STATFS_NODELAY;
2308 rc = obd_statfs(NULL, sbi->ll_md_exp, osfs, max_age, flags);
2312 osfs->os_type = LL_SUPER_MAGIC;
2314 CDEBUG(D_SUPER, "MDC blocks %llu/%llu objects %llu/%llu\n",
2315 osfs->os_bavail, osfs->os_blocks, osfs->os_ffree, osfs->os_files);
2317 if (osfs->os_state & OS_STATFS_SUM)
2320 rc = obd_statfs(NULL, sbi->ll_dt_exp, &obd_osfs, max_age, flags);
2321 if (rc) /* Possibly a filesystem with no OSTs. Report MDT totals. */
2324 CDEBUG(D_SUPER, "OSC blocks %llu/%llu objects %llu/%llu\n",
2325 obd_osfs.os_bavail, obd_osfs.os_blocks, obd_osfs.os_ffree,
2328 osfs->os_bsize = obd_osfs.os_bsize;
2329 osfs->os_blocks = obd_osfs.os_blocks;
2330 osfs->os_bfree = obd_osfs.os_bfree;
2331 osfs->os_bavail = obd_osfs.os_bavail;
2333 /* If we have _some_ OSTs, but don't have as many free objects on the
2334 * OSTs as inodes on the MDTs, reduce the reported number of inodes
2335 * to compensate, so that the "inodes in use" number is correct.
2336 * This should be kept in sync with lod_statfs() behaviour.
2338 if (obd_osfs.os_files && obd_osfs.os_ffree < osfs->os_ffree) {
2339 osfs->os_files = (osfs->os_files - osfs->os_ffree) +
2341 osfs->os_ffree = obd_osfs.os_ffree;
2348 static int ll_statfs_project(struct inode *inode, struct kstatfs *sfs)
2350 struct if_quotactl qctl = {
2351 .qc_cmd = LUSTRE_Q_GETQUOTA,
2352 .qc_type = PRJQUOTA,
2353 .qc_valid = QC_GENERAL,
2355 u64 limit, curblock;
2358 qctl.qc_id = ll_i2info(inode)->lli_projid;
2359 ret = quotactl_ioctl(inode->i_sb, &qctl);
2361 /* ignore errors if project ID does not have
2362 * a quota limit or feature unsupported.
2364 if (ret == -ESRCH || ret == -EOPNOTSUPP)
2369 limit = ((qctl.qc_dqblk.dqb_bsoftlimit ?
2370 qctl.qc_dqblk.dqb_bsoftlimit :
2371 qctl.qc_dqblk.dqb_bhardlimit) * 1024) / sfs->f_bsize;
2372 if (limit && sfs->f_blocks > limit) {
2373 curblock = (qctl.qc_dqblk.dqb_curspace +
2374 sfs->f_bsize - 1) / sfs->f_bsize;
2375 sfs->f_blocks = limit;
2376 sfs->f_bfree = sfs->f_bavail =
2377 (sfs->f_blocks > curblock) ?
2378 (sfs->f_blocks - curblock) : 0;
2381 limit = qctl.qc_dqblk.dqb_isoftlimit ?
2382 qctl.qc_dqblk.dqb_isoftlimit :
2383 qctl.qc_dqblk.dqb_ihardlimit;
2384 if (limit && sfs->f_files > limit) {
2385 sfs->f_files = limit;
2386 sfs->f_ffree = (sfs->f_files >
2387 qctl.qc_dqblk.dqb_curinodes) ?
2388 (sfs->f_files - qctl.qc_dqblk.dqb_curinodes) : 0;
2394 int ll_statfs(struct dentry *de, struct kstatfs *sfs)
2396 struct super_block *sb = de->d_sb;
2397 struct obd_statfs osfs;
2398 __u64 fsid = huge_encode_dev(sb->s_dev);
2399 ktime_t kstart = ktime_get();
2402 CDEBUG(D_VFSTRACE, "VFS Op:sb=%s (%p)\n", sb->s_id, sb);
2404 /* Some amount of caching on the client is allowed */
2405 rc = ll_statfs_internal(ll_s2sbi(sb), &osfs, OBD_STATFS_SUM);
2409 statfs_unpack(sfs, &osfs);
2411 /* We need to downshift for all 32-bit kernels, because we can't
2412 * tell if the kernel is being called via sys_statfs64() or not.
2413 * Stop before overflowing f_bsize - in which case it is better
2414 * to just risk EOVERFLOW if caller is using old sys_statfs(). */
2415 if (sizeof(long) < 8) {
2416 while (osfs.os_blocks > ~0UL && sfs->f_bsize < 0x40000000) {
2419 osfs.os_blocks >>= 1;
2420 osfs.os_bfree >>= 1;
2421 osfs.os_bavail >>= 1;
2425 sfs->f_blocks = osfs.os_blocks;
2426 sfs->f_bfree = osfs.os_bfree;
2427 sfs->f_bavail = osfs.os_bavail;
2428 sfs->f_fsid.val[0] = (__u32)fsid;
2429 sfs->f_fsid.val[1] = (__u32)(fsid >> 32);
2430 if (ll_i2info(de->d_inode)->lli_projid)
2431 return ll_statfs_project(de->d_inode, sfs);
2433 ll_stats_ops_tally(ll_s2sbi(sb), LPROC_LL_STATFS,
2434 ktime_us_delta(ktime_get(), kstart));
2439 void ll_inode_size_lock(struct inode *inode)
2441 struct ll_inode_info *lli;
2443 LASSERT(!S_ISDIR(inode->i_mode));
2445 lli = ll_i2info(inode);
2446 mutex_lock(&lli->lli_size_mutex);
2449 void ll_inode_size_unlock(struct inode *inode)
2451 struct ll_inode_info *lli;
2453 lli = ll_i2info(inode);
2454 mutex_unlock(&lli->lli_size_mutex);
2457 void ll_update_inode_flags(struct inode *inode, unsigned int ext_flags)
2459 /* do not clear encryption flag */
2460 ext_flags |= ll_inode_to_ext_flags(inode->i_flags) & LUSTRE_ENCRYPT_FL;
2461 inode->i_flags = ll_ext_to_inode_flags(ext_flags);
2462 if (ext_flags & LUSTRE_PROJINHERIT_FL)
2463 set_bit(LLIF_PROJECT_INHERIT, &ll_i2info(inode)->lli_flags);
2465 clear_bit(LLIF_PROJECT_INHERIT, &ll_i2info(inode)->lli_flags);
2468 int ll_update_inode(struct inode *inode, struct lustre_md *md)
2470 struct ll_inode_info *lli = ll_i2info(inode);
2471 struct mdt_body *body = md->body;
2472 struct ll_sb_info *sbi = ll_i2sbi(inode);
2476 if (body->mbo_valid & OBD_MD_FLEASIZE) {
2477 rc = cl_file_inode_init(inode, md);
2482 if (S_ISDIR(inode->i_mode)) {
2483 rc = ll_update_lsm_md(inode, md);
2488 if (body->mbo_valid & OBD_MD_FLACL)
2489 lli_replace_acl(lli, md);
2491 api32 = test_bit(LL_SBI_32BIT_API, sbi->ll_flags);
2492 inode->i_ino = cl_fid_build_ino(&body->mbo_fid1, api32);
2493 inode->i_generation = cl_fid_build_gen(&body->mbo_fid1);
2495 if (body->mbo_valid & OBD_MD_FLATIME) {
2496 if (body->mbo_atime > inode->i_atime.tv_sec)
2497 inode->i_atime.tv_sec = body->mbo_atime;
2498 lli->lli_atime = body->mbo_atime;
2501 if (body->mbo_valid & OBD_MD_FLMTIME) {
2502 if (body->mbo_mtime > inode->i_mtime.tv_sec) {
2504 "setting ino %lu mtime from %lld to %llu\n",
2505 inode->i_ino, (s64)inode->i_mtime.tv_sec,
2507 inode->i_mtime.tv_sec = body->mbo_mtime;
2509 lli->lli_mtime = body->mbo_mtime;
2512 if (body->mbo_valid & OBD_MD_FLCTIME) {
2513 if (body->mbo_ctime > inode->i_ctime.tv_sec)
2514 inode->i_ctime.tv_sec = body->mbo_ctime;
2515 lli->lli_ctime = body->mbo_ctime;
2518 if (body->mbo_valid & OBD_MD_FLBTIME)
2519 lli->lli_btime = body->mbo_btime;
2521 /* Clear i_flags to remove S_NOSEC before permissions are updated */
2522 if (body->mbo_valid & OBD_MD_FLFLAGS)
2523 ll_update_inode_flags(inode, body->mbo_flags);
2524 if (body->mbo_valid & OBD_MD_FLMODE)
2525 inode->i_mode = (inode->i_mode & S_IFMT) |
2526 (body->mbo_mode & ~S_IFMT);
2528 if (body->mbo_valid & OBD_MD_FLTYPE)
2529 inode->i_mode = (inode->i_mode & ~S_IFMT) |
2530 (body->mbo_mode & S_IFMT);
2532 LASSERT(inode->i_mode != 0);
2533 if (body->mbo_valid & OBD_MD_FLUID)
2534 inode->i_uid = make_kuid(&init_user_ns, body->mbo_uid);
2535 if (body->mbo_valid & OBD_MD_FLGID)
2536 inode->i_gid = make_kgid(&init_user_ns, body->mbo_gid);
2537 if (body->mbo_valid & OBD_MD_FLPROJID)
2538 lli->lli_projid = body->mbo_projid;
2539 if (body->mbo_valid & OBD_MD_FLNLINK) {
2540 spin_lock(&inode->i_lock);
2541 set_nlink(inode, body->mbo_nlink);
2542 spin_unlock(&inode->i_lock);
2544 if (body->mbo_valid & OBD_MD_FLRDEV)
2545 inode->i_rdev = old_decode_dev(body->mbo_rdev);
2547 if (body->mbo_valid & OBD_MD_FLID) {
2548 /* FID shouldn't be changed! */
2549 if (fid_is_sane(&lli->lli_fid)) {
2550 LASSERTF(lu_fid_eq(&lli->lli_fid, &body->mbo_fid1),
2551 "Trying to change FID "DFID
2552 " to the "DFID", inode "DFID"(%p)\n",
2553 PFID(&lli->lli_fid), PFID(&body->mbo_fid1),
2554 PFID(ll_inode2fid(inode)), inode);
2556 lli->lli_fid = body->mbo_fid1;
2560 LASSERT(fid_seq(&lli->lli_fid) != 0);
2562 lli->lli_attr_valid = body->mbo_valid;
2563 if (body->mbo_valid & OBD_MD_FLSIZE) {
2564 i_size_write(inode, body->mbo_size);
2566 CDEBUG(D_VFSTRACE, "inode="DFID", updating i_size %llu\n",
2567 PFID(ll_inode2fid(inode)),
2568 (unsigned long long)body->mbo_size);
2570 if (body->mbo_valid & OBD_MD_FLBLOCKS)
2571 inode->i_blocks = body->mbo_blocks;
2573 if (body->mbo_valid & OBD_MD_FLLAZYSIZE)
2574 lli->lli_lazysize = body->mbo_size;
2575 if (body->mbo_valid & OBD_MD_FLLAZYBLOCKS)
2576 lli->lli_lazyblocks = body->mbo_blocks;
2579 if (body->mbo_valid & OBD_MD_TSTATE) {
2580 /* Set LLIF_FILE_RESTORING if restore ongoing and
2581 * clear it when done to ensure to start again
2582 * glimpsing updated attrs
2584 if (body->mbo_t_state & MS_RESTORE)
2585 set_bit(LLIF_FILE_RESTORING, &lli->lli_flags);
2587 clear_bit(LLIF_FILE_RESTORING, &lli->lli_flags);
2593 /* update directory depth to ROOT, called after LOOKUP lock is fetched. */
2594 void ll_update_dir_depth(struct inode *dir, struct inode *inode)
2596 struct ll_inode_info *lli;
2598 if (!S_ISDIR(inode->i_mode))
2604 lli = ll_i2info(inode);
2605 lli->lli_depth = ll_i2info(dir)->lli_depth + 1;
2606 CDEBUG(D_INODE, DFID" depth %hu\n", PFID(&lli->lli_fid), lli->lli_depth);
2609 void ll_truncate_inode_pages_final(struct inode *inode)
2611 struct address_space *mapping = &inode->i_data;
2612 unsigned long nrpages;
2613 unsigned long flags;
2615 truncate_inode_pages_final(mapping);
2617 /* Workaround for LU-118: Note nrpages may not be totally updated when
2618 * truncate_inode_pages() returns, as there can be a page in the process
2619 * of deletion (inside __delete_from_page_cache()) in the specified
2620 * range. Thus mapping->nrpages can be non-zero when this function
2621 * returns even after truncation of the whole mapping. Only do this if
2622 * npages isn't already zero.
2624 nrpages = mapping->nrpages;
2626 ll_xa_lock_irqsave(&mapping->i_pages, flags);
2627 nrpages = mapping->nrpages;
2628 ll_xa_unlock_irqrestore(&mapping->i_pages, flags);
2629 } /* Workaround end */
2631 LASSERTF(nrpages == 0, "%s: inode="DFID"(%p) nrpages=%lu, "
2632 "see https://jira.whamcloud.com/browse/LU-118\n",
2633 ll_i2sbi(inode)->ll_fsname,
2634 PFID(ll_inode2fid(inode)), inode, nrpages);
2637 int ll_read_inode2(struct inode *inode, void *opaque)
2639 struct lustre_md *md = opaque;
2640 struct ll_inode_info *lli = ll_i2info(inode);
2644 CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p)\n",
2645 PFID(&lli->lli_fid), inode);
2647 /* Core attributes from the MDS first. This is a new inode, and
2648 * the VFS doesn't zero times in the core inode so we have to do
2649 * it ourselves. They will be overwritten by either MDS or OST
2650 * attributes - we just need to make sure they aren't newer.
2652 inode->i_mtime.tv_sec = 0;
2653 inode->i_atime.tv_sec = 0;
2654 inode->i_ctime.tv_sec = 0;
2656 rc = ll_update_inode(inode, md);
2660 /* OIDEBUG(inode); */
2662 #ifdef HAVE_BACKING_DEV_INFO
2663 /* initializing backing dev info. */
2664 inode->i_mapping->backing_dev_info = &s2lsi(inode->i_sb)->lsi_bdi;
2666 if (S_ISREG(inode->i_mode)) {
2667 struct ll_sb_info *sbi = ll_i2sbi(inode);
2668 inode->i_op = &ll_file_inode_operations;
2669 inode->i_fop = sbi->ll_fop;
2670 inode->i_mapping->a_ops = (struct address_space_operations *)&ll_aops;
2672 } else if (S_ISDIR(inode->i_mode)) {
2673 inode->i_op = &ll_dir_inode_operations;
2674 inode->i_fop = &ll_dir_operations;
2676 } else if (S_ISLNK(inode->i_mode)) {
2677 inode->i_op = &ll_fast_symlink_inode_operations;
2680 inode->i_op = &ll_special_inode_operations;
2682 init_special_inode(inode, inode->i_mode,
2691 void ll_delete_inode(struct inode *inode)
2693 struct ll_inode_info *lli = ll_i2info(inode);
2696 if (S_ISREG(inode->i_mode) && lli->lli_clob != NULL) {
2697 /* It is last chance to write out dirty pages,
2698 * otherwise we may lose data while umount.
2700 * If i_nlink is 0 then just discard data. This is safe because
2701 * local inode gets i_nlink 0 from server only for the last
2702 * unlink, so that file is not opened somewhere else
2704 cl_sync_file_range(inode, 0, OBD_OBJECT_EOF, inode->i_nlink ?
2705 CL_FSYNC_LOCAL : CL_FSYNC_DISCARD, 1);
2708 ll_truncate_inode_pages_final(inode);
2709 ll_clear_inode(inode);
2715 int ll_iocontrol(struct inode *inode, struct file *file,
2716 unsigned int cmd, unsigned long arg)
2718 struct ll_sb_info *sbi = ll_i2sbi(inode);
2719 struct ptlrpc_request *req = NULL;
2724 case FS_IOC_GETFLAGS: {
2725 struct mdt_body *body;
2726 struct md_op_data *op_data;
2728 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL,
2729 0, 0, LUSTRE_OPC_ANY,
2731 if (IS_ERR(op_data))
2732 RETURN(PTR_ERR(op_data));
2734 op_data->op_valid = OBD_MD_FLFLAGS;
2735 rc = md_getattr(sbi->ll_md_exp, op_data, &req);
2736 ll_finish_md_op_data(op_data);
2738 CERROR("%s: failure inode "DFID": rc = %d\n",
2739 sbi->ll_md_exp->exp_obd->obd_name,
2740 PFID(ll_inode2fid(inode)), rc);
2744 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
2746 flags = body->mbo_flags;
2748 ptlrpc_req_finished(req);
2750 RETURN(put_user(flags, (int __user *)arg));
2752 case FS_IOC_SETFLAGS: {
2754 struct md_op_data *op_data;
2755 struct cl_object *obj;
2756 struct fsxattr fa = { 0 };
2758 if (get_user(flags, (int __user *)arg))
2761 fa.fsx_projid = ll_i2info(inode)->lli_projid;
2762 if (flags & LUSTRE_PROJINHERIT_FL)
2763 fa.fsx_xflags = FS_XFLAG_PROJINHERIT;
2765 rc = ll_ioctl_check_project(inode, fa.fsx_xflags,
2770 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
2771 LUSTRE_OPC_ANY, NULL);
2772 if (IS_ERR(op_data))
2773 RETURN(PTR_ERR(op_data));
2775 op_data->op_attr_flags = flags;
2776 op_data->op_xvalid |= OP_XVALID_FLAGS;
2777 rc = md_setattr(sbi->ll_md_exp, op_data, NULL, 0, &req);
2778 ll_finish_md_op_data(op_data);
2779 ptlrpc_req_finished(req);
2783 ll_update_inode_flags(inode, flags);
2785 obj = ll_i2info(inode)->lli_clob;
2789 OBD_ALLOC_PTR(attr);
2793 rc = cl_setattr_ost(obj, attr, OP_XVALID_FLAGS, flags);
2805 int ll_flush_ctx(struct inode *inode)
2807 struct ll_sb_info *sbi = ll_i2sbi(inode);
2809 CDEBUG(D_SEC, "flush context for user %d\n",
2810 from_kuid(&init_user_ns, current_uid()));
2812 obd_set_info_async(NULL, sbi->ll_md_exp,
2813 sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
2815 obd_set_info_async(NULL, sbi->ll_dt_exp,
2816 sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
2821 /* umount -f client means force down, don't save state */
2822 void ll_umount_begin(struct super_block *sb)
2824 struct ll_sb_info *sbi = ll_s2sbi(sb);
2825 struct obd_device *obd;
2826 struct obd_ioctl_data *ioc_data;
2830 CDEBUG(D_VFSTRACE, "VFS Op: superblock %p count %d active %d\n", sb,
2831 sb->s_count, atomic_read(&sb->s_active));
2833 obd = class_exp2obd(sbi->ll_md_exp);
2835 CERROR("Invalid MDC connection handle %#llx\n",
2836 sbi->ll_md_exp->exp_handle.h_cookie);
2842 obd = class_exp2obd(sbi->ll_dt_exp);
2844 CERROR("Invalid LOV connection handle %#llx\n",
2845 sbi->ll_dt_exp->exp_handle.h_cookie);
2851 OBD_ALLOC_PTR(ioc_data);
2853 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_md_exp,
2854 sizeof *ioc_data, ioc_data, NULL);
2856 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_dt_exp,
2857 sizeof *ioc_data, ioc_data, NULL);
2859 OBD_FREE_PTR(ioc_data);
2862 /* Really, we'd like to wait until there are no requests outstanding,
2863 * and then continue. For now, we just periodically checking for vfs
2864 * to decrement mnt_cnt and hope to finish it within 10sec.
2868 !may_umount(sbi->ll_mnt.mnt)) {
2876 int ll_remount_fs(struct super_block *sb, int *flags, char *data)
2878 struct ll_sb_info *sbi = ll_s2sbi(sb);
2879 char *profilenm = get_profile_name(sb);
2883 if ((*flags & MS_RDONLY) != (sb->s_flags & SB_RDONLY)) {
2884 read_only = *flags & MS_RDONLY;
2885 err = obd_set_info_async(NULL, sbi->ll_md_exp,
2886 sizeof(KEY_READ_ONLY),
2887 KEY_READ_ONLY, sizeof(read_only),
2890 LCONSOLE_WARN("Failed to remount %s %s (%d)\n",
2891 profilenm, read_only ?
2892 "read-only" : "read-write", err);
2897 sb->s_flags |= SB_RDONLY;
2899 sb->s_flags &= ~SB_RDONLY;
2901 if (test_bit(LL_SBI_VERBOSE, sbi->ll_flags))
2902 LCONSOLE_WARN("Remounted %s %s\n", profilenm,
2903 read_only ? "read-only" : "read-write");
2909 * Cleanup the open handle that is cached on MDT-side.
2911 * For open case, the client side open handling thread may hit error
2912 * after the MDT grant the open. Under such case, the client should
2913 * send close RPC to the MDT as cleanup; otherwise, the open handle
2914 * on the MDT will be leaked there until the client umount or evicted.
2916 * In further, if someone unlinked the file, because the open handle
2917 * holds the reference on such file/object, then it will block the
2918 * subsequent threads that want to locate such object via FID.
2920 * \param[in] sb super block for this file-system
2921 * \param[in] open_req pointer to the original open request
2923 void ll_open_cleanup(struct super_block *sb, struct req_capsule *pill)
2925 struct mdt_body *body;
2926 struct md_op_data *op_data;
2927 struct ptlrpc_request *close_req = NULL;
2928 struct obd_export *exp = ll_s2sbi(sb)->ll_md_exp;
2931 body = req_capsule_server_get(pill, &RMF_MDT_BODY);
2932 OBD_ALLOC_PTR(op_data);
2933 if (op_data == NULL) {
2934 CWARN("%s: cannot allocate op_data to release open handle for "
2935 DFID"\n", ll_s2sbi(sb)->ll_fsname, PFID(&body->mbo_fid1));
2940 op_data->op_fid1 = body->mbo_fid1;
2941 op_data->op_open_handle = body->mbo_open_handle;
2942 op_data->op_mod_time = ktime_get_real_seconds();
2943 md_close(exp, op_data, NULL, &close_req);
2944 ptlrpc_req_finished(close_req);
2945 ll_finish_md_op_data(op_data);
2950 int ll_prep_inode(struct inode **inode, struct req_capsule *pill,
2951 struct super_block *sb, struct lookup_intent *it)
2953 struct ll_sb_info *sbi = NULL;
2954 struct lustre_md md = { NULL };
2955 bool default_lmv_deleted = false;
2960 LASSERT(*inode || sb);
2961 sbi = sb ? ll_s2sbi(sb) : ll_i2sbi(*inode);
2962 rc = md_get_lustre_md(sbi->ll_md_exp, pill, sbi->ll_dt_exp,
2963 sbi->ll_md_exp, &md);
2968 * clear default_lmv only if intent_getattr reply doesn't contain it.
2969 * but it needs to be done after iget, check this early because
2970 * ll_update_lsm_md() may change md.
2972 if (it && (it->it_op & (IT_LOOKUP | IT_GETATTR)) &&
2973 S_ISDIR(md.body->mbo_mode) && !md.default_lmv)
2974 default_lmv_deleted = true;
2977 rc = ll_update_inode(*inode, &md);
2981 bool api32 = test_bit(LL_SBI_32BIT_API, sbi->ll_flags);
2982 struct lu_fid *fid1 = &md.body->mbo_fid1;
2984 LASSERT(sb != NULL);
2987 * At this point server returns to client's same fid as client
2988 * generated for creating. So using ->fid1 is okay here.
2990 if (!fid_is_sane(fid1)) {
2991 CERROR("%s: Fid is insane "DFID"\n",
2992 sbi->ll_fsname, PFID(fid1));
2993 GOTO(out, rc = -EINVAL);
2996 *inode = ll_iget(sb, cl_fid_build_ino(fid1, api32), &md);
2997 if (IS_ERR(*inode)) {
2999 rc = IS_ERR(*inode) ? PTR_ERR(*inode) : -ENOMEM;
3001 CERROR("new_inode -fatal: rc %d\n", rc);
3006 /* Handling piggyback layout lock.
3007 * Layout lock can be piggybacked by getattr and open request.
3008 * The lsm can be applied to inode only if it comes with a layout lock
3009 * otherwise correct layout may be overwritten, for example:
3010 * 1. proc1: mdt returns a lsm but not granting layout
3011 * 2. layout was changed by another client
3012 * 3. proc2: refresh layout and layout lock granted
3013 * 4. proc1: to apply a stale layout */
3014 if (it != NULL && it->it_lock_mode != 0) {
3015 struct lustre_handle lockh;
3016 struct ldlm_lock *lock;
3018 lockh.cookie = it->it_lock_handle;
3019 lock = ldlm_handle2lock(&lockh);
3020 LASSERT(lock != NULL);
3021 if (ldlm_has_layout(lock)) {
3022 struct cl_object_conf conf;
3024 memset(&conf, 0, sizeof(conf));
3025 conf.coc_opc = OBJECT_CONF_SET;
3026 conf.coc_inode = *inode;
3027 conf.coc_lock = lock;
3028 conf.u.coc_layout = md.layout;
3029 (void)ll_layout_conf(*inode, &conf);
3031 LDLM_LOCK_PUT(lock);
3034 if (default_lmv_deleted)
3035 ll_update_default_lsm_md(*inode, &md);
3037 /* we may want to apply some policy for foreign file/dir */
3038 if (ll_sbi_has_foreign_symlink(sbi)) {
3039 rc = ll_manage_foreign(*inode, &md);
3047 /* cleanup will be done if necessary */
3048 md_free_lustre_md(sbi->ll_md_exp, &md);
3050 if (rc != 0 && it != NULL && it->it_op & IT_OPEN) {
3051 ll_intent_drop_lock(it);
3052 ll_open_cleanup(sb != NULL ? sb : (*inode)->i_sb, pill);
3058 int ll_obd_statfs(struct inode *inode, void __user *arg)
3060 struct ll_sb_info *sbi = NULL;
3061 struct obd_export *exp;
3062 struct obd_ioctl_data *data = NULL;
3067 sbi = ll_i2sbi(inode);
3069 GOTO(out_statfs, rc = -EINVAL);
3071 rc = obd_ioctl_getdata(&data, &len, arg);
3073 GOTO(out_statfs, rc);
3075 if (!data->ioc_inlbuf1 || !data->ioc_inlbuf2 ||
3076 !data->ioc_pbuf1 || !data->ioc_pbuf2)
3077 GOTO(out_statfs, rc = -EINVAL);
3079 if (data->ioc_inllen1 != sizeof(__u32) ||
3080 data->ioc_inllen2 != sizeof(__u32) ||
3081 data->ioc_plen1 != sizeof(struct obd_statfs) ||
3082 data->ioc_plen2 != sizeof(struct obd_uuid))
3083 GOTO(out_statfs, rc = -EINVAL);
3085 memcpy(&type, data->ioc_inlbuf1, sizeof(__u32));
3086 if (type & LL_STATFS_LMV)
3087 exp = sbi->ll_md_exp;
3088 else if (type & LL_STATFS_LOV)
3089 exp = sbi->ll_dt_exp;
3091 GOTO(out_statfs, rc = -ENODEV);
3093 rc = obd_iocontrol(IOC_OBD_STATFS, exp, len, data, NULL);
3095 GOTO(out_statfs, rc);
3097 OBD_FREE_LARGE(data, len);
3102 * this is normally called in ll_fini_md_op_data(), but sometimes it needs to
3103 * be called early to avoid deadlock.
3105 void ll_unlock_md_op_lsm(struct md_op_data *op_data)
3107 if (op_data->op_mea2_sem) {
3108 up_read_non_owner(op_data->op_mea2_sem);
3109 op_data->op_mea2_sem = NULL;
3112 if (op_data->op_mea1_sem) {
3113 up_read_non_owner(op_data->op_mea1_sem);
3114 op_data->op_mea1_sem = NULL;
3118 /* this function prepares md_op_data hint for passing it down to MD stack. */
3119 struct md_op_data *ll_prep_md_op_data(struct md_op_data *op_data,
3120 struct inode *i1, struct inode *i2,
3121 const char *name, size_t namelen,
3122 __u32 mode, enum md_op_code opc,
3125 struct llcrypt_name fname = { 0 };
3128 LASSERT(i1 != NULL);
3131 /* Do not reuse namelen for something else. */
3133 return ERR_PTR(-EINVAL);
3135 if (namelen > ll_i2sbi(i1)->ll_namelen)
3136 return ERR_PTR(-ENAMETOOLONG);
3138 /* "/" is not valid name, but it's allowed */
3139 if (!lu_name_is_valid_2(name, namelen) &&
3140 strncmp("/", name, namelen) != 0)
3141 return ERR_PTR(-EINVAL);
3144 if (op_data == NULL)
3145 OBD_ALLOC_PTR(op_data);
3147 if (op_data == NULL)
3148 return ERR_PTR(-ENOMEM);
3150 ll_i2gids(op_data->op_suppgids, i1, i2);
3151 op_data->op_fid1 = *ll_inode2fid(i1);
3153 if (S_ISDIR(i1->i_mode)) {
3154 down_read_non_owner(&ll_i2info(i1)->lli_lsm_sem);
3155 op_data->op_mea1_sem = &ll_i2info(i1)->lli_lsm_sem;
3156 op_data->op_mea1 = ll_i2info(i1)->lli_lsm_md;
3157 op_data->op_default_mea1 = ll_i2info(i1)->lli_default_lsm_md;
3161 op_data->op_fid2 = *ll_inode2fid(i2);
3162 if (S_ISDIR(i2->i_mode)) {
3164 /* i2 is typically a child of i1, and MUST be
3165 * further from the root to avoid deadlocks.
3167 down_read_non_owner(&ll_i2info(i2)->lli_lsm_sem);
3168 op_data->op_mea2_sem =
3169 &ll_i2info(i2)->lli_lsm_sem;
3171 op_data->op_mea2 = ll_i2info(i2)->lli_lsm_md;
3174 fid_zero(&op_data->op_fid2);
3177 if (test_bit(LL_SBI_64BIT_HASH, ll_i2sbi(i1)->ll_flags))
3178 op_data->op_cli_flags |= CLI_HASH64;
3180 if (ll_need_32bit_api(ll_i2sbi(i1)))
3181 op_data->op_cli_flags |= CLI_API32;
3183 if (opc == LUSTRE_OPC_LOOKUP || opc == LUSTRE_OPC_CREATE) {
3184 /* In case of lookup, ll_setup_filename() has already been
3185 * called in ll_lookup_it(), so just take provided name.
3187 fname.disk_name.name = (unsigned char *)name;
3188 fname.disk_name.len = namelen;
3189 } else if (name && namelen) {
3190 struct qstr dname = QSTR_INIT(name, namelen);
3192 struct lu_fid *pfid = NULL;
3196 if (!S_ISDIR(i1->i_mode) && i2 && S_ISDIR(i2->i_mode)) {
3197 /* special case when called from ll_link() */
3202 lookup = (int)(opc == LUSTRE_OPC_ANY);
3204 if (opc == LUSTRE_OPC_ANY && lookup)
3206 rc = ll_setup_filename(dir, &dname, lookup, &fname, pfid);
3208 ll_finish_md_op_data(op_data);
3211 if (pfid && !fid_is_zero(pfid)) {
3213 op_data->op_fid2 = fid;
3214 op_data->op_bias = MDS_FID_OP;
3216 if (fname.disk_name.name &&
3217 fname.disk_name.name != (unsigned char *)name)
3218 /* op_data->op_name must be freed after use */
3219 op_data->op_flags |= MF_OPNAME_KMALLOCED;
3222 /* In fact LUSTRE_OPC_LOOKUP, LUSTRE_OPC_OPEN, LUSTRE_OPC_MIGR
3223 * are LUSTRE_OPC_ANY
3225 if (opc == LUSTRE_OPC_LOOKUP || opc == LUSTRE_OPC_OPEN ||
3226 opc == LUSTRE_OPC_MIGR)
3227 op_data->op_code = LUSTRE_OPC_ANY;
3229 op_data->op_code = opc;
3230 op_data->op_name = fname.disk_name.name;
3231 op_data->op_namelen = fname.disk_name.len;
3232 op_data->op_mode = mode;
3233 op_data->op_mod_time = ktime_get_real_seconds();
3234 op_data->op_fsuid = from_kuid(&init_user_ns, current_fsuid());
3235 op_data->op_fsgid = from_kgid(&init_user_ns, current_fsgid());
3236 op_data->op_cap = current_cap();
3237 op_data->op_mds = 0;
3238 if ((opc == LUSTRE_OPC_CREATE) && (name != NULL) &&
3239 filename_is_volatile(name, namelen, &op_data->op_mds)) {
3240 op_data->op_bias |= MDS_CREATE_VOLATILE;
3242 op_data->op_data = data;
3247 void ll_finish_md_op_data(struct md_op_data *op_data)
3249 ll_unlock_md_op_lsm(op_data);
3250 ll_security_release_secctx(op_data->op_file_secctx,
3251 op_data->op_file_secctx_size);
3252 if (op_data->op_flags & MF_OPNAME_KMALLOCED)
3253 /* allocated via ll_setup_filename called
3254 * from ll_prep_md_op_data
3256 kfree(op_data->op_name);
3257 llcrypt_free_ctx(op_data->op_file_encctx, op_data->op_file_encctx_size);
3258 OBD_FREE_PTR(op_data);
3261 int ll_show_options(struct seq_file *seq, struct dentry *dentry)
3263 struct ll_sb_info *sbi;
3266 LASSERT(seq && dentry);
3267 sbi = ll_s2sbi(dentry->d_sb);
3269 if (test_bit(LL_SBI_NOLCK, sbi->ll_flags))
3270 seq_puts(seq, "nolock");
3272 for (i = 1; ll_sbi_flags_name[i].token != LL_SBI_NUM_MOUNT_OPT; i++) {
3273 /* match_table in some cases has patterns for both enabled and
3274 * disabled cases. Ignore 'no'xxx versions if bit is set.
3276 if (test_bit(ll_sbi_flags_name[i].token, sbi->ll_flags) &&
3277 strncmp(ll_sbi_flags_name[i].pattern, "no", 2)) {
3278 if (ll_sbi_flags_name[i].token ==
3279 LL_SBI_FOREIGN_SYMLINK) {
3280 seq_show_option(seq, "foreign_symlink",
3281 sbi->ll_foreign_symlink_prefix);
3283 seq_printf(seq, ",%s",
3284 ll_sbi_flags_name[i].pattern);
3287 /* You can have either localflock or flock but not
3288 * both. If localflock is set don't print flock or
3291 if (ll_sbi_flags_name[i].token == LL_SBI_LOCALFLOCK)
3293 } else if (!test_bit(ll_sbi_flags_name[i].token, sbi->ll_flags) &&
3294 !strncmp(ll_sbi_flags_name[i].pattern, "no", 2)) {
3295 seq_printf(seq, ",%s",
3296 ll_sbi_flags_name[i].pattern);
3304 * Get obd name by cmd, and copy out to user space
3306 int ll_get_obd_name(struct inode *inode, unsigned int cmd, unsigned long arg)
3308 struct ll_sb_info *sbi = ll_i2sbi(inode);
3309 struct obd_device *obd;
3312 if (cmd == OBD_IOC_GETNAME_OLD || cmd == OBD_IOC_GETDTNAME)
3313 obd = class_exp2obd(sbi->ll_dt_exp);
3314 else if (cmd == OBD_IOC_GETMDNAME)
3315 obd = class_exp2obd(sbi->ll_md_exp);
3322 if (copy_to_user((void __user *)arg, obd->obd_name,
3323 strlen(obd->obd_name) + 1))
3329 static char* ll_d_path(struct dentry *dentry, char *buf, int bufsize)
3336 p.mnt = current->fs->root.mnt;
3338 path = d_path(&p, buf, bufsize);
3343 void ll_dirty_page_discard_warn(struct page *page, int ioret)
3345 char *buf, *path = NULL;
3346 struct dentry *dentry = NULL;
3347 struct inode *inode = page->mapping->host;
3349 /* this can be called inside spin lock so use GFP_ATOMIC. */
3350 buf = (char *)__get_free_page(GFP_ATOMIC);
3352 dentry = d_find_alias(page->mapping->host);
3354 path = ll_d_path(dentry, buf, PAGE_SIZE);
3357 /* The below message is checked in recovery-small.sh test_24b */
3359 "%s: dirty page discard: %s/fid: "DFID"/%s may get corrupted "
3360 "(rc %d)\n", ll_i2sbi(inode)->ll_fsname,
3361 s2lsi(page->mapping->host->i_sb)->lsi_lmd->lmd_dev,
3362 PFID(ll_inode2fid(inode)),
3363 (path && !IS_ERR(path)) ? path : "", ioret);
3369 free_page((unsigned long)buf);
3372 ssize_t ll_copy_user_md(const struct lov_user_md __user *md,
3373 struct lov_user_md **kbuf)
3375 struct lov_user_md lum;
3379 if (copy_from_user(&lum, md, sizeof(lum)))
3382 lum_size = ll_lov_user_md_size(&lum);
3386 OBD_ALLOC_LARGE(*kbuf, lum_size);
3390 if (copy_from_user(*kbuf, md, lum_size) != 0) {
3391 OBD_FREE_LARGE(*kbuf, lum_size);
3399 * Compute llite root squash state after a change of root squash
3400 * configuration setting or add/remove of a lnet nid
3402 void ll_compute_rootsquash_state(struct ll_sb_info *sbi)
3404 struct root_squash_info *squash = &sbi->ll_squash;
3407 struct lnet_process_id id;
3409 /* Update norootsquash flag */
3410 spin_lock(&squash->rsi_lock);
3411 if (list_empty(&squash->rsi_nosquash_nids))
3412 clear_bit(LL_SBI_NOROOTSQUASH, sbi->ll_flags);
3414 /* Do not apply root squash as soon as one of our NIDs is
3415 * in the nosquash_nids list */
3418 while (LNetGetId(i++, &id) != -ENOENT) {
3419 if (id.nid == LNET_NID_LO_0)
3421 if (cfs_match_nid(id.nid, &squash->rsi_nosquash_nids)) {
3427 set_bit(LL_SBI_NOROOTSQUASH, sbi->ll_flags);
3429 clear_bit(LL_SBI_NOROOTSQUASH, sbi->ll_flags);
3431 spin_unlock(&squash->rsi_lock);
3435 * Parse linkea content to extract information about a given hardlink
3437 * \param[in] ldata - Initialized linkea data
3438 * \param[in] linkno - Link identifier
3439 * \param[out] parent_fid - The entry's parent FID
3440 * \param[out] ln - Entry name destination buffer
3442 * \retval 0 on success
3443 * \retval Appropriate negative error code on failure
3445 static int ll_linkea_decode(struct linkea_data *ldata, unsigned int linkno,
3446 struct lu_fid *parent_fid, struct lu_name *ln)
3452 rc = linkea_init_with_rec(ldata);
3456 if (linkno >= ldata->ld_leh->leh_reccount)
3457 /* beyond last link */
3460 linkea_first_entry(ldata);
3461 for (idx = 0; ldata->ld_lee != NULL; idx++) {
3462 linkea_entry_unpack(ldata->ld_lee, &ldata->ld_reclen, ln,
3467 linkea_next_entry(ldata);
3477 * Get parent FID and name of an identified link. Operation is performed for
3478 * a given link number, letting the caller iterate over linkno to list one or
3479 * all links of an entry.
3481 * \param[in] file - File descriptor against which to perform the operation
3482 * \param[in,out] arg - User-filled structure containing the linkno to operate
3483 * on and the available size. It is eventually filled with
3484 * the requested information or left untouched on error
3486 * \retval - 0 on success
3487 * \retval - Appropriate negative error code on failure
3489 int ll_getparent(struct file *file, struct getparent __user *arg)
3491 struct inode *inode = file_inode(file);
3492 struct linkea_data *ldata;
3493 struct lu_buf buf = LU_BUF_NULL;
3495 struct lu_fid parent_fid;
3502 if (!capable(CAP_DAC_READ_SEARCH) &&
3503 !test_bit(LL_SBI_USER_FID2PATH, ll_i2sbi(inode)->ll_flags))
3506 if (get_user(name_size, &arg->gp_name_size))
3509 if (get_user(linkno, &arg->gp_linkno))
3512 if (name_size > PATH_MAX)
3515 OBD_ALLOC(ldata, sizeof(*ldata));
3519 rc = linkea_data_new(ldata, &buf);
3521 GOTO(ldata_free, rc);
3523 rc = ll_xattr_list(inode, XATTR_NAME_LINK, XATTR_TRUSTED_T, buf.lb_buf,
3524 buf.lb_len, OBD_MD_FLXATTR);
3528 rc = ll_linkea_decode(ldata, linkno, &parent_fid, &ln);
3532 if (ln.ln_namelen >= name_size)
3533 GOTO(lb_free, rc = -EOVERFLOW);
3535 if (copy_to_user(&arg->gp_fid, &parent_fid, sizeof(arg->gp_fid)))
3536 GOTO(lb_free, rc = -EFAULT);
3538 if (copy_to_user(&arg->gp_name, ln.ln_name, ln.ln_namelen))
3539 GOTO(lb_free, rc = -EFAULT);
3541 if (put_user('\0', arg->gp_name + ln.ln_namelen))
3542 GOTO(lb_free, rc = -EFAULT);
3547 OBD_FREE(ldata, sizeof(*ldata));