4 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
6 * This program is free software; you can redistribute it and/or modify
7 * it under the terms of the GNU General Public License version 2 only,
8 * as published by the Free Software Foundation.
10 * This program is distributed in the hope that it will be useful, but
11 * WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
13 * General Public License version 2 for more details (a copy is included
14 * in the LICENSE file that accompanied this code).
16 * You should have received a copy of the GNU General Public License
17 * version 2 along with this program; If not, see
18 * http://www.gnu.org/licenses/gpl-2.0.html
23 * Copyright (c) 2003, 2010, Oracle and/or its affiliates. All rights reserved.
24 * Use is subject to license terms.
26 * Copyright (c) 2011, 2017, Intel Corporation.
29 * This file is part of Lustre, http://www.lustre.org/
31 * lustre/llite/llite_lib.c
33 * Lustre Light Super operations
36 #define DEBUG_SUBSYSTEM S_LLITE
38 #include <linux/cpu.h>
39 #include <linux/module.h>
40 #include <linux/random.h>
41 #include <linux/statfs.h>
42 #include <linux/time.h>
43 #include <linux/file.h>
44 #include <linux/types.h>
45 #include <libcfs/linux/linux-uuid.h>
46 #include <linux/version.h>
48 #include <linux/user_namespace.h>
49 #include <linux/delay.h>
50 #include <linux/uidgid.h>
51 #include <linux/fs_struct.h>
53 #ifndef HAVE_CPUS_READ_LOCK
54 #include <libcfs/linux/linux-cpu.h>
56 #include <libcfs/linux/linux-misc.h>
57 #include <uapi/linux/lustre/lustre_ioctl.h>
58 #ifdef HAVE_UAPI_LINUX_MOUNT_H
59 #include <uapi/linux/mount.h>
62 #include <lustre_ha.h>
63 #include <lustre_dlm.h>
64 #include <lprocfs_status.h>
65 #include <lustre_disk.h>
66 #include <uapi/linux/lustre/lustre_param.h>
67 #include <lustre_log.h>
68 #include <cl_object.h>
69 #include <obd_cksum.h>
70 #include "llite_internal.h"
72 struct kmem_cache *ll_file_data_slab;
75 #define log2(n) ffz(~(n))
79 * If there is only one number of core visible to Lustre,
80 * async readahead will be disabled, to avoid massive over
81 * subscription, we use 1/2 of active cores as default max
82 * async readahead requests.
84 static inline unsigned int ll_get_ra_async_max_active(void)
86 return cfs_cpt_weight(cfs_cpt_tab, CFS_CPT_ANY) >> 1;
89 static struct ll_sb_info *ll_init_sbi(void)
91 struct ll_sb_info *sbi = NULL;
93 unsigned long lru_page_max;
101 RETURN(ERR_PTR(-ENOMEM));
103 rc = pcc_super_init(&sbi->ll_pcc_super);
107 spin_lock_init(&sbi->ll_lock);
108 mutex_init(&sbi->ll_lco.lco_lock);
109 spin_lock_init(&sbi->ll_pp_extent_lock);
110 spin_lock_init(&sbi->ll_process_lock);
111 sbi->ll_rw_stats_on = 0;
112 sbi->ll_statfs_max_age = OBD_STATFS_CACHE_SECONDS;
115 pages = si.totalram - si.totalhigh;
116 lru_page_max = pages / 2;
118 sbi->ll_ra_info.ra_async_max_active = ll_get_ra_async_max_active();
119 sbi->ll_ra_info.ll_readahead_wq =
120 cfs_cpt_bind_workqueue("ll-readahead-wq", cfs_cpt_tab,
122 sbi->ll_ra_info.ra_async_max_active);
123 if (IS_ERR(sbi->ll_ra_info.ll_readahead_wq))
124 GOTO(out_pcc, rc = PTR_ERR(sbi->ll_ra_info.ll_readahead_wq));
126 /* initialize ll_cache data */
127 sbi->ll_cache = cl_cache_init(lru_page_max);
128 if (sbi->ll_cache == NULL)
129 GOTO(out_destroy_ra, rc = -ENOMEM);
131 /* initialize foreign symlink prefix path */
132 OBD_ALLOC(sbi->ll_foreign_symlink_prefix, sizeof("/mnt/"));
133 if (sbi->ll_foreign_symlink_prefix == NULL)
134 GOTO(out_destroy_ra, rc = -ENOMEM);
135 memcpy(sbi->ll_foreign_symlink_prefix, "/mnt/", sizeof("/mnt/"));
136 sbi->ll_foreign_symlink_prefix_size = sizeof("/mnt/");
138 /* initialize foreign symlink upcall path, none by default */
139 OBD_ALLOC(sbi->ll_foreign_symlink_upcall, sizeof("none"));
140 if (sbi->ll_foreign_symlink_upcall == NULL)
141 GOTO(out_destroy_ra, rc = -ENOMEM);
142 memcpy(sbi->ll_foreign_symlink_upcall, "none", sizeof("none"));
143 sbi->ll_foreign_symlink_upcall_items = NULL;
144 sbi->ll_foreign_symlink_upcall_nb_items = 0;
145 init_rwsem(&sbi->ll_foreign_symlink_sem);
146 /* foreign symlink support (LL_SBI_FOREIGN_SYMLINK in ll_flags)
147 * not enabled by default
150 sbi->ll_ra_info.ra_max_pages =
151 min(pages / 32, SBI_DEFAULT_READ_AHEAD_MAX);
152 sbi->ll_ra_info.ra_max_pages_per_file =
153 min(sbi->ll_ra_info.ra_max_pages / 4,
154 SBI_DEFAULT_READ_AHEAD_PER_FILE_MAX);
155 sbi->ll_ra_info.ra_async_pages_per_file_threshold =
156 sbi->ll_ra_info.ra_max_pages_per_file;
157 sbi->ll_ra_info.ra_range_pages = SBI_DEFAULT_RA_RANGE_PAGES;
158 sbi->ll_ra_info.ra_max_read_ahead_whole_pages = -1;
159 atomic_set(&sbi->ll_ra_info.ra_async_inflight, 0);
161 set_bit(LL_SBI_VERBOSE, sbi->ll_flags);
162 #ifdef ENABLE_CHECKSUM
163 set_bit(LL_SBI_CHECKSUM, sbi->ll_flags);
166 set_bit(LL_SBI_FLOCK, sbi->ll_flags);
169 #ifdef HAVE_LRU_RESIZE_SUPPORT
170 set_bit(LL_SBI_LRU_RESIZE, sbi->ll_flags);
172 set_bit(LL_SBI_LAZYSTATFS, sbi->ll_flags);
174 /* metadata statahead is enabled by default */
175 sbi->ll_sa_running_max = LL_SA_RUNNING_DEF;
176 sbi->ll_sa_max = LL_SA_RPC_DEF;
177 atomic_set(&sbi->ll_sa_total, 0);
178 atomic_set(&sbi->ll_sa_wrong, 0);
179 atomic_set(&sbi->ll_sa_running, 0);
180 atomic_set(&sbi->ll_agl_total, 0);
181 set_bit(LL_SBI_AGL_ENABLED, sbi->ll_flags);
182 /* Disable LL_SBI_FAST_READ by default, see LU-15815. */
183 set_bit(LL_SBI_TINY_WRITE, sbi->ll_flags);
184 set_bit(LL_SBI_PARALLEL_DIO, sbi->ll_flags);
185 ll_sbi_set_encrypt(sbi, true);
188 sbi->ll_squash.rsi_uid = 0;
189 sbi->ll_squash.rsi_gid = 0;
190 INIT_LIST_HEAD(&sbi->ll_squash.rsi_nosquash_nids);
191 spin_lock_init(&sbi->ll_squash.rsi_lock);
193 /* Per-filesystem file heat */
194 sbi->ll_heat_decay_weight = SBI_DEFAULT_HEAT_DECAY_WEIGHT;
195 sbi->ll_heat_period_second = SBI_DEFAULT_HEAT_PERIOD_SECOND;
197 /* Per-fs open heat level before requesting open lock */
198 sbi->ll_oc_thrsh_count = SBI_DEFAULT_OPENCACHE_THRESHOLD_COUNT;
199 sbi->ll_oc_max_ms = SBI_DEFAULT_OPENCACHE_THRESHOLD_MAX_MS;
200 sbi->ll_oc_thrsh_ms = SBI_DEFAULT_OPENCACHE_THRESHOLD_MS;
203 if (sbi->ll_foreign_symlink_prefix)
204 OBD_FREE(sbi->ll_foreign_symlink_prefix, sizeof("/mnt/"));
206 cl_cache_decref(sbi->ll_cache);
207 sbi->ll_cache = NULL;
209 destroy_workqueue(sbi->ll_ra_info.ll_readahead_wq);
211 pcc_super_fini(&sbi->ll_pcc_super);
217 static void ll_free_sbi(struct super_block *sb)
219 struct ll_sb_info *sbi = ll_s2sbi(sb);
223 if (!list_empty(&sbi->ll_squash.rsi_nosquash_nids))
224 cfs_free_nidlist(&sbi->ll_squash.rsi_nosquash_nids);
225 if (sbi->ll_ra_info.ll_readahead_wq)
226 destroy_workqueue(sbi->ll_ra_info.ll_readahead_wq);
227 if (sbi->ll_cache != NULL) {
228 cl_cache_decref(sbi->ll_cache);
229 sbi->ll_cache = NULL;
231 if (sbi->ll_foreign_symlink_prefix) {
232 OBD_FREE(sbi->ll_foreign_symlink_prefix,
233 sbi->ll_foreign_symlink_prefix_size);
234 sbi->ll_foreign_symlink_prefix = NULL;
236 if (sbi->ll_foreign_symlink_upcall) {
237 OBD_FREE(sbi->ll_foreign_symlink_upcall,
238 strlen(sbi->ll_foreign_symlink_upcall) +
240 sbi->ll_foreign_symlink_upcall = NULL;
242 if (sbi->ll_foreign_symlink_upcall_items) {
244 int nb_items = sbi->ll_foreign_symlink_upcall_nb_items;
245 struct ll_foreign_symlink_upcall_item *items =
246 sbi->ll_foreign_symlink_upcall_items;
248 for (i = 0 ; i < nb_items; i++)
249 if (items[i].type == STRING_TYPE)
250 OBD_FREE(items[i].string,
253 OBD_FREE_LARGE(items, nb_items *
254 sizeof(struct ll_foreign_symlink_upcall_item));
255 sbi->ll_foreign_symlink_upcall_items = NULL;
257 ll_free_rw_stats_info(sbi);
258 pcc_super_fini(&sbi->ll_pcc_super);
259 OBD_FREE(sbi, sizeof(*sbi));
264 static int client_common_fill_super(struct super_block *sb, char *md, char *dt)
266 struct inode *root = NULL;
267 struct ll_sb_info *sbi = ll_s2sbi(sb);
268 struct obd_statfs *osfs = NULL;
269 struct ptlrpc_request *request = NULL;
270 struct obd_connect_data *data = NULL;
271 struct obd_uuid *uuid;
272 struct md_op_data *op_data;
273 struct lustre_md lmd;
275 int size, err, checksum;
281 sbi->ll_md_obd = class_name2obd(md);
282 if (!sbi->ll_md_obd) {
283 CERROR("MD %s: not setup or attached\n", md);
297 /* pass client page size via ocd_grant_blkbits, the server should report
298 * back its backend blocksize for grant calculation purpose */
299 data->ocd_grant_blkbits = PAGE_SHIFT;
301 /* indicate MDT features supported by this client */
302 data->ocd_connect_flags = OBD_CONNECT_IBITS | OBD_CONNECT_NODEVOH |
303 OBD_CONNECT_ATTRFID | OBD_CONNECT_GRANT |
304 OBD_CONNECT_VERSION | OBD_CONNECT_BRW_SIZE |
305 OBD_CONNECT_SRVLOCK |
306 OBD_CONNECT_MDS_CAPA | OBD_CONNECT_OSS_CAPA |
307 OBD_CONNECT_CANCELSET | OBD_CONNECT_FID |
308 OBD_CONNECT_AT | OBD_CONNECT_LOV_V3 |
309 OBD_CONNECT_VBR | OBD_CONNECT_FULL20 |
310 OBD_CONNECT_64BITHASH |
311 OBD_CONNECT_EINPROGRESS |
312 OBD_CONNECT_JOBSTATS | OBD_CONNECT_LVB_TYPE |
313 OBD_CONNECT_LAYOUTLOCK | OBD_CONNECT_PINGLESS|
314 OBD_CONNECT_MAX_EASIZE |
315 OBD_CONNECT_FLOCK_DEAD |
316 OBD_CONNECT_DISP_STRIPE | OBD_CONNECT_LFSCK |
317 OBD_CONNECT_OPEN_BY_FID |
318 OBD_CONNECT_DIR_STRIPE |
319 OBD_CONNECT_BULK_MBITS | OBD_CONNECT_CKSUM |
320 OBD_CONNECT_SUBTREE |
321 OBD_CONNECT_MULTIMODRPCS |
322 OBD_CONNECT_GRANT_PARAM |
323 OBD_CONNECT_GRANT_SHRINK |
324 OBD_CONNECT_SHORTIO | OBD_CONNECT_FLAGS2;
326 data->ocd_connect_flags2 = OBD_CONNECT2_DIR_MIGRATE |
327 OBD_CONNECT2_SUM_STATFS |
328 OBD_CONNECT2_OVERSTRIPING |
330 OBD_CONNECT2_LOCK_CONVERT |
331 OBD_CONNECT2_ARCHIVE_ID_ARRAY |
332 OBD_CONNECT2_INC_XID |
334 OBD_CONNECT2_ASYNC_DISCARD |
336 OBD_CONNECT2_CRUSH | OBD_CONNECT2_LSEEK |
337 OBD_CONNECT2_GETATTR_PFID |
338 OBD_CONNECT2_DOM_LVB |
339 OBD_CONNECT2_REP_MBITS |
340 OBD_CONNECT2_ATOMIC_OPEN_LOCK;
342 #ifdef HAVE_LRU_RESIZE_SUPPORT
343 if (test_bit(LL_SBI_LRU_RESIZE, sbi->ll_flags))
344 data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
346 data->ocd_connect_flags |= OBD_CONNECT_ACL_FLAGS;
348 data->ocd_cksum_types = obd_cksum_types_supported_client();
350 if (OBD_FAIL_CHECK(OBD_FAIL_MDC_LIGHTWEIGHT))
351 /* flag mdc connection as lightweight, only used for test
352 * purpose, use with care */
353 data->ocd_connect_flags |= OBD_CONNECT_LIGHTWEIGHT;
355 data->ocd_ibits_known = MDS_INODELOCK_FULL;
356 data->ocd_version = LUSTRE_VERSION_CODE;
358 if (sb->s_flags & SB_RDONLY)
359 data->ocd_connect_flags |= OBD_CONNECT_RDONLY;
360 if (test_bit(LL_SBI_USER_XATTR, sbi->ll_flags))
361 data->ocd_connect_flags |= OBD_CONNECT_XATTR;
364 /* Setting this indicates we correctly support S_NOSEC (See kernel
365 * commit 9e1f1de02c2275d7172e18dc4e7c2065777611bf)
367 sb->s_flags |= SB_NOSEC;
369 sbi->ll_fop = ll_select_file_operations(sbi);
371 /* always ping even if server suppress_pings */
372 if (test_bit(LL_SBI_ALWAYS_PING, sbi->ll_flags))
373 data->ocd_connect_flags &= ~OBD_CONNECT_PINGLESS;
375 obd_connect_set_secctx(data);
376 if (ll_sbi_has_encrypt(sbi))
377 obd_connect_set_enc(data);
379 #if defined(CONFIG_SECURITY)
380 data->ocd_connect_flags2 |= OBD_CONNECT2_SELINUX_POLICY;
383 data->ocd_brw_size = MD_MAX_BRW_SIZE;
385 err = obd_connect(NULL, &sbi->ll_md_exp, sbi->ll_md_obd,
386 &sbi->ll_sb_uuid, data, sbi->ll_cache);
388 LCONSOLE_ERROR_MSG(0x14f, "An MDT (md %s) is performing "
389 "recovery, of which this client is not a "
390 "part. Please wait for recovery to complete,"
391 " abort, or time out.\n", md);
394 CERROR("cannot connect to %s: rc = %d\n", md, err);
398 sbi->ll_md_exp->exp_connect_data = *data;
400 err = obd_fid_init(sbi->ll_md_exp->exp_obd, sbi->ll_md_exp,
401 LUSTRE_SEQ_METADATA);
403 CERROR("%s: Can't init metadata layer FID infrastructure, "
404 "rc = %d\n", sbi->ll_md_exp->exp_obd->obd_name, err);
408 /* For mount, we only need fs info from MDT0, and also in DNE, it
409 * can make sure the client can be mounted as long as MDT0 is
411 err = obd_statfs(NULL, sbi->ll_md_exp, osfs,
412 ktime_get_seconds() - sbi->ll_statfs_max_age,
413 OBD_STATFS_FOR_MDT0);
415 GOTO(out_md_fid, err);
417 /* This needs to be after statfs to ensure connect has finished.
418 * Note that "data" does NOT contain the valid connect reply.
419 * If connecting to a 1.8 server there will be no LMV device, so
420 * we can access the MDC export directly and exp_connect_flags will
421 * be non-zero, but if accessing an upgraded 2.1 server it will
422 * have the correct flags filled in.
423 * XXX: fill in the LMV exp_connect_flags from MDC(s). */
424 valid = exp_connect_flags(sbi->ll_md_exp) & CLIENT_CONNECT_MDT_REQD;
425 if (exp_connect_flags(sbi->ll_md_exp) != 0 &&
426 valid != CLIENT_CONNECT_MDT_REQD) {
429 OBD_ALLOC_WAIT(buf, PAGE_SIZE);
430 obd_connect_flags2str(buf, PAGE_SIZE,
431 valid ^ CLIENT_CONNECT_MDT_REQD, 0, ",");
432 LCONSOLE_ERROR_MSG(0x170, "Server %s does not support "
433 "feature(s) needed for correct operation "
434 "of this client (%s). Please upgrade "
435 "server or downgrade client.\n",
436 sbi->ll_md_exp->exp_obd->obd_name, buf);
437 OBD_FREE(buf, PAGE_SIZE);
438 GOTO(out_md_fid, err = -EPROTO);
441 size = sizeof(*data);
442 err = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_CONN_DATA),
443 KEY_CONN_DATA, &size, data);
445 CERROR("%s: Get connect data failed: rc = %d\n",
446 sbi->ll_md_exp->exp_obd->obd_name, err);
447 GOTO(out_md_fid, err);
450 LASSERT(osfs->os_bsize);
451 sb->s_blocksize = osfs->os_bsize;
452 sb->s_blocksize_bits = log2(osfs->os_bsize);
453 sb->s_magic = LL_SUPER_MAGIC;
454 sb->s_maxbytes = MAX_LFS_FILESIZE;
455 sbi->ll_namelen = osfs->os_namelen;
456 sbi->ll_mnt.mnt = current->fs->root.mnt;
457 sbi->ll_mnt_ns = current->nsproxy->mnt_ns;
459 if (test_bit(LL_SBI_USER_XATTR, sbi->ll_flags) &&
460 !(data->ocd_connect_flags & OBD_CONNECT_XATTR)) {
461 LCONSOLE_INFO("Disabling user_xattr feature because "
462 "it is not supported on the server\n");
463 clear_bit(LL_SBI_USER_XATTR, sbi->ll_flags);
466 if (data->ocd_connect_flags & OBD_CONNECT_ACL) {
468 sb->s_flags |= SB_POSIXACL;
470 set_bit(LL_SBI_ACL, sbi->ll_flags);
472 LCONSOLE_INFO("client wants to enable acl, but mdt not!\n");
474 sb->s_flags &= ~SB_POSIXACL;
476 clear_bit(LL_SBI_ACL, sbi->ll_flags);
479 if (data->ocd_connect_flags & OBD_CONNECT_64BITHASH)
480 set_bit(LL_SBI_64BIT_HASH, sbi->ll_flags);
482 if (data->ocd_connect_flags & OBD_CONNECT_LAYOUTLOCK)
483 set_bit(LL_SBI_LAYOUT_LOCK, sbi->ll_flags);
485 if (obd_connect_has_secctx(data))
486 set_bit(LL_SBI_FILE_SECCTX, sbi->ll_flags);
488 if (ll_sbi_has_encrypt(sbi) && !obd_connect_has_enc(data)) {
489 if (ll_sbi_has_test_dummy_encryption(sbi))
490 LCONSOLE_WARN("%s: server %s does not support encryption feature, encryption deactivated.\n",
492 sbi->ll_md_exp->exp_obd->obd_name);
493 ll_sbi_set_encrypt(sbi, false);
496 if (data->ocd_ibits_known & MDS_INODELOCK_XATTR) {
497 if (!(data->ocd_connect_flags & OBD_CONNECT_MAX_EASIZE)) {
498 LCONSOLE_INFO("%s: disabling xattr cache due to "
499 "unknown maximum xattr size.\n", dt);
500 } else if (!sbi->ll_xattr_cache_set) {
501 /* If xattr_cache is already set (no matter 0 or 1)
502 * during processing llog, it won't be enabled here. */
503 set_bit(LL_SBI_XATTR_CACHE, sbi->ll_flags);
504 sbi->ll_xattr_cache_enabled = 1;
508 sbi->ll_dt_obd = class_name2obd(dt);
509 if (!sbi->ll_dt_obd) {
510 CERROR("DT %s: not setup or attached\n", dt);
511 GOTO(out_md_fid, err = -ENODEV);
514 /* pass client page size via ocd_grant_blkbits, the server should report
515 * back its backend blocksize for grant calculation purpose */
516 data->ocd_grant_blkbits = PAGE_SHIFT;
518 /* indicate OST features supported by this client */
519 data->ocd_connect_flags = OBD_CONNECT_GRANT | OBD_CONNECT_VERSION |
520 OBD_CONNECT_REQPORTAL | OBD_CONNECT_BRW_SIZE |
521 OBD_CONNECT_CANCELSET | OBD_CONNECT_FID |
522 OBD_CONNECT_SRVLOCK |
523 OBD_CONNECT_AT | OBD_CONNECT_OSS_CAPA |
524 OBD_CONNECT_VBR | OBD_CONNECT_FULL20 |
525 OBD_CONNECT_64BITHASH | OBD_CONNECT_MAXBYTES |
526 OBD_CONNECT_EINPROGRESS |
527 OBD_CONNECT_JOBSTATS | OBD_CONNECT_LVB_TYPE |
528 OBD_CONNECT_LAYOUTLOCK |
529 OBD_CONNECT_PINGLESS | OBD_CONNECT_LFSCK |
530 OBD_CONNECT_BULK_MBITS | OBD_CONNECT_SHORTIO |
531 OBD_CONNECT_FLAGS2 | OBD_CONNECT_GRANT_SHRINK;
532 data->ocd_connect_flags2 = OBD_CONNECT2_LOCKAHEAD |
533 OBD_CONNECT2_INC_XID | OBD_CONNECT2_LSEEK |
534 OBD_CONNECT2_REP_MBITS;
536 if (!OBD_FAIL_CHECK(OBD_FAIL_OSC_CONNECT_GRANT_PARAM))
537 data->ocd_connect_flags |= OBD_CONNECT_GRANT_PARAM;
539 /* OBD_CONNECT_CKSUM should always be set, even if checksums are
540 * disabled by default, because it can still be enabled on the
541 * fly via /sys. As a consequence, we still need to come to an
542 * agreement on the supported algorithms at connect time
544 data->ocd_connect_flags |= OBD_CONNECT_CKSUM;
546 if (OBD_FAIL_CHECK(OBD_FAIL_OSC_CKSUM_ADLER_ONLY))
547 data->ocd_cksum_types = OBD_CKSUM_ADLER;
549 data->ocd_cksum_types = obd_cksum_types_supported_client();
551 #ifdef HAVE_LRU_RESIZE_SUPPORT
552 data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
554 /* always ping even if server suppress_pings */
555 if (test_bit(LL_SBI_ALWAYS_PING, sbi->ll_flags))
556 data->ocd_connect_flags &= ~OBD_CONNECT_PINGLESS;
558 if (ll_sbi_has_encrypt(sbi))
559 obd_connect_set_enc(data);
561 CDEBUG(D_RPCTRACE, "ocd_connect_flags: %#llx ocd_version: %d "
562 "ocd_grant: %d\n", data->ocd_connect_flags,
563 data->ocd_version, data->ocd_grant);
565 sbi->ll_dt_obd->obd_upcall.onu_owner = &sbi->ll_lco;
566 sbi->ll_dt_obd->obd_upcall.onu_upcall = cl_ocd_update;
568 data->ocd_brw_size = DT_MAX_BRW_SIZE;
570 err = obd_connect(NULL, &sbi->ll_dt_exp, sbi->ll_dt_obd,
571 &sbi->ll_sb_uuid, data, sbi->ll_cache);
573 LCONSOLE_ERROR_MSG(0x150, "An OST (dt %s) is performing "
574 "recovery, of which this client is not a "
575 "part. Please wait for recovery to "
576 "complete, abort, or time out.\n", dt);
579 CERROR("%s: Cannot connect to %s: rc = %d\n",
580 sbi->ll_dt_exp->exp_obd->obd_name, dt, err);
584 if (ll_sbi_has_encrypt(sbi) &&
585 !obd_connect_has_enc(&sbi->ll_dt_obd->u.lov.lov_ocd)) {
586 if (ll_sbi_has_test_dummy_encryption(sbi))
587 LCONSOLE_WARN("%s: server %s does not support encryption feature, encryption deactivated.\n",
589 ll_sbi_set_encrypt(sbi, false);
590 } else if (ll_sbi_has_test_dummy_encryption(sbi)) {
591 LCONSOLE_WARN("Test dummy encryption mode enabled\n");
594 sbi->ll_dt_exp->exp_connect_data = *data;
596 /* Don't change value if it was specified in the config log */
597 if (sbi->ll_ra_info.ra_max_read_ahead_whole_pages == -1) {
598 sbi->ll_ra_info.ra_max_read_ahead_whole_pages =
599 max_t(unsigned long, SBI_DEFAULT_READ_AHEAD_WHOLE_MAX,
600 (data->ocd_brw_size >> PAGE_SHIFT));
601 if (sbi->ll_ra_info.ra_max_read_ahead_whole_pages >
602 sbi->ll_ra_info.ra_max_pages_per_file)
603 sbi->ll_ra_info.ra_max_read_ahead_whole_pages =
604 sbi->ll_ra_info.ra_max_pages_per_file;
607 err = obd_fid_init(sbi->ll_dt_exp->exp_obd, sbi->ll_dt_exp,
608 LUSTRE_SEQ_METADATA);
610 CERROR("%s: Can't init data layer FID infrastructure, "
611 "rc = %d\n", sbi->ll_dt_exp->exp_obd->obd_name, err);
615 mutex_lock(&sbi->ll_lco.lco_lock);
616 sbi->ll_lco.lco_flags = data->ocd_connect_flags;
617 sbi->ll_lco.lco_md_exp = sbi->ll_md_exp;
618 sbi->ll_lco.lco_dt_exp = sbi->ll_dt_exp;
619 mutex_unlock(&sbi->ll_lco.lco_lock);
621 fid_zero(&sbi->ll_root_fid);
622 err = md_get_root(sbi->ll_md_exp, get_mount_fileset(sb),
625 CERROR("cannot mds_connect: rc = %d\n", err);
626 GOTO(out_lock_cn_cb, err);
628 if (!fid_is_sane(&sbi->ll_root_fid)) {
629 CERROR("%s: Invalid root fid "DFID" during mount\n",
630 sbi->ll_md_exp->exp_obd->obd_name,
631 PFID(&sbi->ll_root_fid));
632 GOTO(out_lock_cn_cb, err = -EINVAL);
634 CDEBUG(D_SUPER, "rootfid "DFID"\n", PFID(&sbi->ll_root_fid));
636 sb->s_op = &lustre_super_operations;
637 sb->s_xattr = ll_xattr_handlers;
638 #if THREAD_SIZE >= 8192 /*b=17630*/
639 sb->s_export_op = &lustre_export_operations;
641 #ifdef HAVE_LUSTRE_CRYPTO
642 llcrypt_set_ops(sb, &lustre_cryptops);
646 * XXX: move this to after cbd setup? */
647 valid = OBD_MD_FLGETATTR | OBD_MD_FLBLOCKS | OBD_MD_FLMODEASIZE |
649 if (test_bit(LL_SBI_ACL, sbi->ll_flags))
650 valid |= OBD_MD_FLACL;
652 OBD_ALLOC_PTR(op_data);
654 GOTO(out_lock_cn_cb, err = -ENOMEM);
656 op_data->op_fid1 = sbi->ll_root_fid;
657 op_data->op_mode = 0;
658 op_data->op_valid = valid;
660 err = md_getattr(sbi->ll_md_exp, op_data, &request);
662 /* We need enc ctx info, so reset it in op_data to
663 * prevent it from being freed.
665 encctx = op_data->op_file_encctx;
666 encctxlen = op_data->op_file_encctx_size;
667 op_data->op_file_encctx = NULL;
668 op_data->op_file_encctx_size = 0;
669 OBD_FREE_PTR(op_data);
671 CERROR("%s: md_getattr failed for root: rc = %d\n",
672 sbi->ll_md_exp->exp_obd->obd_name, err);
673 GOTO(out_lock_cn_cb, err);
676 err = md_get_lustre_md(sbi->ll_md_exp, &request->rq_pill,
677 sbi->ll_dt_exp, sbi->ll_md_exp, &lmd);
679 CERROR("failed to understand root inode md: rc = %d\n", err);
680 ptlrpc_req_finished(request);
681 GOTO(out_lock_cn_cb, err);
684 LASSERT(fid_is_sane(&sbi->ll_root_fid));
685 api32 = test_bit(LL_SBI_32BIT_API, sbi->ll_flags);
686 root = ll_iget(sb, cl_fid_build_ino(&sbi->ll_root_fid, api32), &lmd);
687 md_free_lustre_md(sbi->ll_md_exp, &lmd);
691 err = IS_ERR(root) ? PTR_ERR(root) : -EBADF;
693 CERROR("%s: bad ll_iget() for root: rc = %d\n",
694 sbi->ll_fsname, err);
695 ptlrpc_req_finished(request);
701 "server returned encryption ctx for root inode "DFID"\n",
702 PFID(&sbi->ll_root_fid));
703 err = ll_set_encflags(root, encctx, encctxlen, true);
705 CWARN("%s: cannot set enc ctx for "DFID": rc = %d\n",
707 PFID(&sbi->ll_root_fid), err);
709 ptlrpc_req_finished(request);
711 checksum = test_bit(LL_SBI_CHECKSUM, sbi->ll_flags);
712 if (sbi->ll_checksum_set) {
713 err = obd_set_info_async(NULL, sbi->ll_dt_exp,
714 sizeof(KEY_CHECKSUM), KEY_CHECKSUM,
715 sizeof(checksum), &checksum, NULL);
717 CERROR("%s: Set checksum failed: rc = %d\n",
718 sbi->ll_dt_exp->exp_obd->obd_name, err);
724 sb->s_root = d_make_root(root);
725 if (sb->s_root == NULL) {
727 CERROR("%s: can't make root dentry: rc = %d\n",
728 sbi->ll_fsname, err);
732 sbi->ll_sdev_orig = sb->s_dev;
734 /* We set sb->s_dev equal on all lustre clients in order to support
735 * NFS export clustering. NFSD requires that the FSID be the same
737 /* s_dev is also used in lt_compare() to compare two fs, but that is
738 * only a node-local comparison. */
739 uuid = obd_get_uuid(sbi->ll_md_exp);
741 sb->s_dev = get_uuid2int(uuid->uuid, strlen(uuid->uuid));
748 if (sbi->ll_dt_obd) {
749 err = sysfs_create_link(&sbi->ll_kset.kobj,
750 &sbi->ll_dt_obd->obd_kset.kobj,
751 sbi->ll_dt_obd->obd_type->typ_name);
753 CERROR("%s: could not register %s in llite: rc = %d\n",
754 dt, sbi->ll_fsname, err);
759 if (sbi->ll_md_obd) {
760 err = sysfs_create_link(&sbi->ll_kset.kobj,
761 &sbi->ll_md_obd->obd_kset.kobj,
762 sbi->ll_md_obd->obd_type->typ_name);
764 CERROR("%s: could not register %s in llite: rc = %d\n",
765 md, sbi->ll_fsname, err);
774 obd_fid_fini(sbi->ll_dt_exp->exp_obd);
776 obd_disconnect(sbi->ll_dt_exp);
777 sbi->ll_dt_exp = NULL;
778 sbi->ll_dt_obd = NULL;
780 obd_fid_fini(sbi->ll_md_exp->exp_obd);
782 obd_disconnect(sbi->ll_md_exp);
783 sbi->ll_md_exp = NULL;
784 sbi->ll_md_obd = NULL;
793 int ll_get_max_mdsize(struct ll_sb_info *sbi, int *lmmsize)
797 size = sizeof(*lmmsize);
798 rc = obd_get_info(NULL, sbi->ll_dt_exp, sizeof(KEY_MAX_EASIZE),
799 KEY_MAX_EASIZE, &size, lmmsize);
801 CERROR("%s: cannot get max LOV EA size: rc = %d\n",
802 sbi->ll_dt_exp->exp_obd->obd_name, rc);
806 CDEBUG(D_INFO, "max LOV ea size: %d\n", *lmmsize);
809 rc = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_MAX_EASIZE),
810 KEY_MAX_EASIZE, &size, lmmsize);
812 CERROR("Get max mdsize error rc %d\n", rc);
814 CDEBUG(D_INFO, "max LMV ea size: %d\n", *lmmsize);
820 * Get the value of the default_easize parameter.
822 * \see client_obd::cl_default_mds_easize
824 * \param[in] sbi superblock info for this filesystem
825 * \param[out] lmmsize pointer to storage location for value
827 * \retval 0 on success
828 * \retval negative negated errno on failure
830 int ll_get_default_mdsize(struct ll_sb_info *sbi, int *lmmsize)
835 rc = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_DEFAULT_EASIZE),
836 KEY_DEFAULT_EASIZE, &size, lmmsize);
838 CERROR("Get default mdsize error rc %d\n", rc);
844 * Set the default_easize parameter to the given value.
846 * \see client_obd::cl_default_mds_easize
848 * \param[in] sbi superblock info for this filesystem
849 * \param[in] lmmsize the size to set
851 * \retval 0 on success
852 * \retval negative negated errno on failure
854 int ll_set_default_mdsize(struct ll_sb_info *sbi, int lmmsize)
858 if (lmmsize < sizeof(struct lov_mds_md) ||
859 lmmsize > OBD_MAX_DEFAULT_EA_SIZE)
862 rc = obd_set_info_async(NULL, sbi->ll_md_exp,
863 sizeof(KEY_DEFAULT_EASIZE), KEY_DEFAULT_EASIZE,
864 sizeof(int), &lmmsize, NULL);
869 static void client_common_put_super(struct super_block *sb)
871 struct ll_sb_info *sbi = ll_s2sbi(sb);
876 obd_fid_fini(sbi->ll_dt_exp->exp_obd);
877 obd_disconnect(sbi->ll_dt_exp);
878 sbi->ll_dt_exp = NULL;
880 ll_debugfs_unregister_super(sb);
882 obd_fid_fini(sbi->ll_md_exp->exp_obd);
883 obd_disconnect(sbi->ll_md_exp);
884 sbi->ll_md_exp = NULL;
889 void ll_kill_super(struct super_block *sb)
891 struct ll_sb_info *sbi;
895 if (!(sb->s_flags & SB_ACTIVE))
899 /* we need restore s_dev from changed for clustred NFS before put_super
900 * because new kernels have cached s_dev and change sb->s_dev in
901 * put_super not affected real removing devices */
903 sb->s_dev = sbi->ll_sdev_orig;
905 /* wait running statahead threads to quit */
906 while (atomic_read(&sbi->ll_sa_running) > 0)
907 schedule_timeout_uninterruptible(
908 cfs_time_seconds(1) >> 3);
914 /* Since we use this table for ll_sbi_flags_seq_show make
915 * sure what you want displayed for a specific token that
916 * is listed more than once below be listed first. For
917 * example we want "checksum" displayed, not "nochecksum"
920 static const match_table_t ll_sbi_flags_name = {
921 {LL_SBI_NOLCK, "nolock"},
922 {LL_SBI_CHECKSUM, "checksum"},
923 {LL_SBI_CHECKSUM, "nochecksum"},
924 {LL_SBI_LOCALFLOCK, "localflock"},
925 {LL_SBI_FLOCK, "flock"},
926 {LL_SBI_FLOCK, "noflock"},
927 {LL_SBI_USER_XATTR, "user_xattr"},
928 {LL_SBI_USER_XATTR, "nouser_xattr"},
929 {LL_SBI_LRU_RESIZE, "lruresize"},
930 {LL_SBI_LRU_RESIZE, "nolruresize"},
931 {LL_SBI_LAZYSTATFS, "lazystatfs"},
932 {LL_SBI_LAZYSTATFS, "nolazystatfs"},
933 {LL_SBI_32BIT_API, "32bitapi"},
934 {LL_SBI_USER_FID2PATH, "user_fid2path"},
935 {LL_SBI_USER_FID2PATH, "nouser_fid2path"},
936 {LL_SBI_VERBOSE, "verbose"},
937 {LL_SBI_VERBOSE, "noverbose"},
938 {LL_SBI_ALWAYS_PING, "always_ping"},
939 {LL_SBI_TEST_DUMMY_ENCRYPTION, "test_dummy_encryption"},
940 {LL_SBI_ENCRYPT, "encrypt"},
941 {LL_SBI_ENCRYPT, "noencrypt"},
942 {LL_SBI_FOREIGN_SYMLINK, "foreign_symlink=%s"},
943 {LL_SBI_NUM_MOUNT_OPT, NULL},
946 {LL_SBI_AGL_ENABLED, "agl"},
947 {LL_SBI_64BIT_HASH, "64bit_hash"},
948 {LL_SBI_LAYOUT_LOCK, "layout"},
949 {LL_SBI_XATTR_CACHE, "xattr_cache"},
950 {LL_SBI_NOROOTSQUASH, "norootsquash"},
951 {LL_SBI_FAST_READ, "fast_read"},
952 {LL_SBI_FILE_SECCTX, "file_secctx"},
953 {LL_SBI_TINY_WRITE, "tiny_write"},
954 {LL_SBI_FILE_HEAT, "file_heat"},
955 {LL_SBI_PARALLEL_DIO, "parallel_dio"},
958 int ll_sbi_flags_seq_show(struct seq_file *m, void *v)
960 struct super_block *sb = m->private;
963 for (i = 0; i < LL_SBI_NUM_FLAGS; i++) {
966 if (!test_bit(i, ll_s2sbi(sb)->ll_flags))
969 for (j = 0; j < ARRAY_SIZE(ll_sbi_flags_name); j++) {
970 if (ll_sbi_flags_name[j].token == i &&
971 ll_sbi_flags_name[j].pattern) {
973 ll_sbi_flags_name[j].pattern);
982 /* non-client-specific mount options are parsed in lmd_parse */
983 static int ll_options(char *options, struct super_block *sb)
985 struct ll_sb_info *sbi = ll_s2sbi(sb);
986 char *s2, *s1, *opts;
992 /* Don't stomp on lmd_opts */
993 opts = kstrdup(options, GFP_KERNEL);
999 CDEBUG(D_CONFIG, "Parsing opts %s\n", options);
1001 while ((s1 = strsep(&opts, ",")) != NULL) {
1002 substring_t args[MAX_OPT_ARGS];
1003 bool turn_off = false;
1009 CDEBUG(D_SUPER, "next opt=%s\n", s1);
1011 if (strncmp(s1, "no", 2) == 0)
1015 * Initialize args struct so we know whether arg was
1016 * found; some options take optional arguments.
1019 args[0].from = NULL;
1020 token = match_token(s1, ll_sbi_flags_name, args);
1021 if (token == LL_SBI_NUM_MOUNT_OPT) {
1022 if (match_wildcard("context", s1) ||
1023 match_wildcard("fscontext", s1) ||
1024 match_wildcard("defcontext", s1) ||
1025 match_wildcard("rootcontext",s1))
1028 LCONSOLE_ERROR_MSG(0x152,
1029 "Unknown option '%s', won't mount.\n",
1036 case LL_SBI_32BIT_API:
1037 case LL_SBI_64BIT_HASH:
1038 case LL_SBI_ALWAYS_PING:
1039 set_bit(token, sbi->ll_flags);
1043 clear_bit(LL_SBI_LOCALFLOCK, sbi->ll_flags);
1045 clear_bit(LL_SBI_FLOCK, sbi->ll_flags);
1047 set_bit(token, sbi->ll_flags);
1050 case LL_SBI_LOCALFLOCK:
1051 clear_bit(LL_SBI_FLOCK, sbi->ll_flags);
1052 set_bit(token, sbi->ll_flags);
1055 case LL_SBI_CHECKSUM:
1056 sbi->ll_checksum_set = 1;
1058 case LL_SBI_USER_XATTR:
1059 case LL_SBI_USER_FID2PATH:
1060 case LL_SBI_LRU_RESIZE:
1061 case LL_SBI_LAZYSTATFS:
1062 case LL_SBI_VERBOSE:
1064 clear_bit(token, sbi->ll_flags);
1066 set_bit(token, sbi->ll_flags);
1068 case LL_SBI_TEST_DUMMY_ENCRYPTION: {
1069 #ifdef HAVE_LUSTRE_CRYPTO
1070 set_bit(token, sbi->ll_flags);
1072 LCONSOLE_WARN("Test dummy encryption mount option ignored: encryption not supported\n");
1076 case LL_SBI_ENCRYPT:
1077 #ifdef HAVE_LUSTRE_CRYPTO
1079 clear_bit(token, sbi->ll_flags);
1081 set_bit(token, sbi->ll_flags);
1083 LCONSOLE_WARN("noencrypt or encrypt mount option ignored: encryption not supported\n");
1086 case LL_SBI_FOREIGN_SYMLINK:
1087 /* non-default prefix provided ? */
1092 /* path must be absolute */
1093 if (args->from[0] != '/') {
1094 LCONSOLE_ERROR_MSG(0x152,
1095 "foreign prefix '%s' must be an absolute path\n",
1100 old_len = sbi->ll_foreign_symlink_prefix_size;
1101 old = sbi->ll_foreign_symlink_prefix;
1102 /* alloc for path length and '\0' */
1103 sbi->ll_foreign_symlink_prefix = match_strdup(args);
1104 if (!sbi->ll_foreign_symlink_prefix) {
1105 /* restore previous */
1106 sbi->ll_foreign_symlink_prefix = old;
1107 sbi->ll_foreign_symlink_prefix_size =
1111 sbi->ll_foreign_symlink_prefix_size =
1112 args->to - args->from + 1;
1113 OBD_ALLOC_POST(sbi->ll_foreign_symlink_prefix,
1114 sbi->ll_foreign_symlink_prefix_size,
1117 OBD_FREE(old, old_len);
1119 /* enable foreign symlink support */
1120 set_bit(token, sbi->ll_flags);
1122 LCONSOLE_ERROR_MSG(0x152,
1123 "invalid %s option\n", s1);
1134 void ll_lli_init(struct ll_inode_info *lli)
1136 lli->lli_inode_magic = LLI_INODE_MAGIC;
1138 rwlock_init(&lli->lli_lock);
1139 lli->lli_posix_acl = NULL;
1140 /* Do not set lli_fid, it has been initialized already. */
1141 fid_zero(&lli->lli_pfid);
1142 lli->lli_mds_read_och = NULL;
1143 lli->lli_mds_write_och = NULL;
1144 lli->lli_mds_exec_och = NULL;
1145 lli->lli_open_fd_read_count = 0;
1146 lli->lli_open_fd_write_count = 0;
1147 lli->lli_open_fd_exec_count = 0;
1148 mutex_init(&lli->lli_och_mutex);
1149 spin_lock_init(&lli->lli_agl_lock);
1150 spin_lock_init(&lli->lli_layout_lock);
1151 ll_layout_version_set(lli, CL_LAYOUT_GEN_NONE);
1152 lli->lli_clob = NULL;
1154 init_rwsem(&lli->lli_xattrs_list_rwsem);
1155 mutex_init(&lli->lli_xattrs_enq_lock);
1157 LASSERT(lli->lli_vfs_inode.i_mode != 0);
1158 if (S_ISDIR(lli->lli_vfs_inode.i_mode)) {
1159 lli->lli_opendir_key = NULL;
1160 lli->lli_sai = NULL;
1161 spin_lock_init(&lli->lli_sa_lock);
1162 lli->lli_opendir_pid = 0;
1163 lli->lli_sa_enabled = 0;
1164 init_rwsem(&lli->lli_lsm_sem);
1166 mutex_init(&lli->lli_size_mutex);
1167 mutex_init(&lli->lli_setattr_mutex);
1168 lli->lli_symlink_name = NULL;
1169 ll_trunc_sem_init(&lli->lli_trunc_sem);
1170 range_lock_tree_init(&lli->lli_write_tree);
1171 init_rwsem(&lli->lli_glimpse_sem);
1172 lli->lli_glimpse_time = ktime_set(0, 0);
1173 INIT_LIST_HEAD(&lli->lli_agl_list);
1174 lli->lli_agl_index = 0;
1175 lli->lli_async_rc = 0;
1176 spin_lock_init(&lli->lli_heat_lock);
1177 obd_heat_clear(lli->lli_heat_instances, OBD_HEAT_COUNT);
1178 lli->lli_heat_flags = 0;
1179 mutex_init(&lli->lli_pcc_lock);
1180 lli->lli_pcc_state = PCC_STATE_FL_NONE;
1181 lli->lli_pcc_inode = NULL;
1182 lli->lli_pcc_dsflags = PCC_DATASET_INVALID;
1183 lli->lli_pcc_generation = 0;
1184 mutex_init(&lli->lli_group_mutex);
1185 lli->lli_group_users = 0;
1186 lli->lli_group_gid = 0;
1188 mutex_init(&lli->lli_layout_mutex);
1189 memset(lli->lli_jobid, 0, sizeof(lli->lli_jobid));
1190 /* ll_cl_context initialize */
1191 INIT_LIST_HEAD(&lli->lli_lccs);
1194 #define MAX_STRING_SIZE 128
1196 #ifndef HAVE_SUPER_SETUP_BDI_NAME
1198 #define LSI_BDI_INITIALIZED 0x00400000
1200 #ifndef HAVE_BDI_CAP_MAP_COPY
1201 # define BDI_CAP_MAP_COPY 0
1204 static int super_setup_bdi_name(struct super_block *sb, char *fmt, ...)
1206 struct lustre_sb_info *lsi = s2lsi(sb);
1207 char buf[MAX_STRING_SIZE];
1211 err = bdi_init(&lsi->lsi_bdi);
1215 lsi->lsi_flags |= LSI_BDI_INITIALIZED;
1216 lsi->lsi_bdi.capabilities = BDI_CAP_MAP_COPY;
1217 lsi->lsi_bdi.name = "lustre";
1218 va_start(args, fmt);
1219 vsnprintf(buf, MAX_STRING_SIZE, fmt, args);
1221 err = bdi_register(&lsi->lsi_bdi, NULL, "%s", buf);
1224 sb->s_bdi = &lsi->lsi_bdi;
1228 #endif /* !HAVE_SUPER_SETUP_BDI_NAME */
1230 int ll_fill_super(struct super_block *sb)
1232 struct lustre_profile *lprof = NULL;
1233 struct lustre_sb_info *lsi = s2lsi(sb);
1234 struct ll_sb_info *sbi = NULL;
1235 char *dt = NULL, *md = NULL;
1236 char *profilenm = get_profile_name(sb);
1237 struct config_llog_instance *cfg;
1238 /* %p for void* in printf needs 16+2 characters: 0xffffffffffffffff */
1239 const int instlen = LUSTRE_MAXINSTANCE + 2;
1240 unsigned long cfg_instance = ll_get_cfg_instance(sb);
1241 char name[MAX_STRING_SIZE];
1250 /* for ASLR, to map between cfg_instance and hashed ptr */
1251 CDEBUG(D_VFSTRACE, "VFS Op: cfg_instance %s-%016lx (sb %p)\n",
1252 profilenm, cfg_instance, sb);
1254 OBD_RACE(OBD_FAIL_LLITE_RACE_MOUNT);
1258 GOTO(out_free_cfg, err = -ENOMEM);
1260 /* client additional sb info */
1261 lsi->lsi_llsbi = sbi = ll_init_sbi();
1263 GOTO(out_free_cfg, err = PTR_ERR(sbi));
1265 err = ll_options(lsi->lsi_lmd->lmd_opts, sb);
1267 GOTO(out_free_cfg, err);
1269 /* kernel >= 2.6.38 store dentry operations in sb->s_d_op. */
1270 sb->s_d_op = &ll_d_ops;
1273 generate_random_uuid(uuid.b);
1274 snprintf(sbi->ll_sb_uuid.uuid, sizeof(sbi->ll_sb_uuid), "%pU", uuid.b);
1276 CDEBUG(D_CONFIG, "llite sb uuid: %s\n", sbi->ll_sb_uuid.uuid);
1279 len = strlen(profilenm);
1280 ptr = strrchr(profilenm, '-');
1281 if (ptr && (strcmp(ptr, "-client") == 0))
1284 if (len > LUSTRE_MAXFSNAME) {
1285 if (unlikely(len >= MAX_STRING_SIZE))
1286 len = MAX_STRING_SIZE - 1;
1287 strncpy(name, profilenm, len);
1289 err = -ENAMETOOLONG;
1290 CERROR("%s: fsname longer than %u characters: rc = %d\n",
1291 name, LUSTRE_MAXFSNAME, err);
1292 GOTO(out_free_cfg, err);
1294 strncpy(sbi->ll_fsname, profilenm, len);
1295 sbi->ll_fsname[len] = '\0';
1298 snprintf(name, sizeof(name), "%.*s-%016lx", len,
1299 profilenm, cfg_instance);
1301 err = super_setup_bdi_name(sb, "%s", name);
1303 GOTO(out_free_cfg, err);
1305 /* disable kernel readahead */
1306 sb->s_bdi->ra_pages = 0;
1308 /* Call ll_debugfs_register_super() before lustre_process_log()
1309 * so that "llite.*.*" params can be processed correctly.
1311 err = ll_debugfs_register_super(sb, name);
1313 CERROR("%s: could not register mountpoint in llite: rc = %d\n",
1314 sbi->ll_fsname, err);
1318 /* The cfg_instance is a value unique to this super, in case some
1319 * joker tries to mount the same fs at two mount points.
1321 cfg->cfg_instance = cfg_instance;
1322 cfg->cfg_uuid = lsi->lsi_llsbi->ll_sb_uuid;
1323 cfg->cfg_callback = class_config_llog_handler;
1324 cfg->cfg_sub_clds = CONFIG_SUB_CLIENT;
1325 /* set up client obds */
1326 err = lustre_process_log(sb, profilenm, cfg);
1328 GOTO(out_debugfs, err);
1330 /* Profile set with LCFG_MOUNTOPT so we can find our mdc and osc obds */
1331 lprof = class_get_profile(profilenm);
1332 if (lprof == NULL) {
1333 LCONSOLE_ERROR_MSG(0x156, "The client profile '%s' could not be"
1334 " read from the MGS. Does that filesystem "
1335 "exist?\n", profilenm);
1336 GOTO(out_debugfs, err = -EINVAL);
1338 CDEBUG(D_CONFIG, "Found profile %s: mdc=%s osc=%s\n", profilenm,
1339 lprof->lp_md, lprof->lp_dt);
1341 dt_len = strlen(lprof->lp_dt) + instlen + 2;
1342 OBD_ALLOC(dt, dt_len);
1344 GOTO(out_profile, err = -ENOMEM);
1345 snprintf(dt, dt_len - 1, "%s-%016lx", lprof->lp_dt, cfg_instance);
1347 md_len = strlen(lprof->lp_md) + instlen + 2;
1348 OBD_ALLOC(md, md_len);
1350 GOTO(out_free_dt, err = -ENOMEM);
1351 snprintf(md, md_len - 1, "%s-%016lx", lprof->lp_md, cfg_instance);
1353 /* connections, registrations, sb setup */
1354 err = client_common_fill_super(sb, md, dt);
1356 GOTO(out_free_md, err);
1358 sbi->ll_client_common_fill_super_succeeded = 1;
1362 OBD_FREE(md, md_len);
1365 OBD_FREE(dt, dt_len);
1368 class_put_profile(lprof);
1371 ll_debugfs_unregister_super(sb);
1378 else if (test_bit(LL_SBI_VERBOSE, sbi->ll_flags))
1379 LCONSOLE_WARN("Mounted %s\n", profilenm);
1381 } /* ll_fill_super */
1383 void ll_put_super(struct super_block *sb)
1385 struct config_llog_instance cfg, params_cfg;
1386 struct obd_device *obd;
1387 struct lustre_sb_info *lsi = s2lsi(sb);
1388 struct ll_sb_info *sbi = ll_s2sbi(sb);
1389 char *profilenm = get_profile_name(sb);
1390 unsigned long cfg_instance = ll_get_cfg_instance(sb);
1392 int next, force = 1, rc = 0;
1396 GOTO(out_no_sbi, 0);
1398 /* Should replace instance_id with something better for ASLR */
1399 CDEBUG(D_VFSTRACE, "VFS Op: cfg_instance %s-%016lx (sb %p)\n",
1400 profilenm, cfg_instance, sb);
1402 cfg.cfg_instance = cfg_instance;
1403 lustre_end_log(sb, profilenm, &cfg);
1405 params_cfg.cfg_instance = cfg_instance;
1406 lustre_end_log(sb, PARAMS_FILENAME, ¶ms_cfg);
1408 if (sbi->ll_md_exp) {
1409 obd = class_exp2obd(sbi->ll_md_exp);
1411 force = obd->obd_force;
1414 /* Wait for unstable pages to be committed to stable storage */
1416 rc = l_wait_event_abortable(
1417 sbi->ll_cache->ccc_unstable_waitq,
1418 atomic_long_read(&sbi->ll_cache->ccc_unstable_nr) == 0);
1421 ccc_count = atomic_long_read(&sbi->ll_cache->ccc_unstable_nr);
1422 if (force == 0 && rc != -ERESTARTSYS)
1423 LASSERTF(ccc_count == 0, "count: %li\n", ccc_count);
1425 /* We need to set force before the lov_disconnect in
1426 * lustre_common_put_super, since l_d cleans up osc's as well.
1430 while ((obd = class_devices_in_group(&sbi->ll_sb_uuid,
1432 obd->obd_force = force;
1436 if (sbi->ll_client_common_fill_super_succeeded) {
1437 /* Only if client_common_fill_super succeeded */
1438 client_common_put_super(sb);
1442 while ((obd = class_devices_in_group(&sbi->ll_sb_uuid, &next)))
1443 class_manual_cleanup(obd);
1445 if (test_bit(LL_SBI_VERBOSE, sbi->ll_flags))
1446 LCONSOLE_WARN("Unmounted %s\n", profilenm ? profilenm : "");
1449 class_del_profile(profilenm);
1451 #ifndef HAVE_SUPER_SETUP_BDI_NAME
1452 if (lsi->lsi_flags & LSI_BDI_INITIALIZED) {
1453 bdi_destroy(&lsi->lsi_bdi);
1454 lsi->lsi_flags &= ~LSI_BDI_INITIALIZED;
1459 lsi->lsi_llsbi = NULL;
1461 lustre_common_put_super(sb);
1463 cl_env_cache_purge(~0);
1466 } /* client_put_super */
1468 struct inode *ll_inode_from_resource_lock(struct ldlm_lock *lock)
1470 struct inode *inode = NULL;
1472 /* NOTE: we depend on atomic igrab() -bzzz */
1473 lock_res_and_lock(lock);
1474 if (lock->l_resource->lr_lvb_inode) {
1475 struct ll_inode_info * lli;
1476 lli = ll_i2info(lock->l_resource->lr_lvb_inode);
1477 if (lli->lli_inode_magic == LLI_INODE_MAGIC) {
1478 inode = igrab(lock->l_resource->lr_lvb_inode);
1480 inode = lock->l_resource->lr_lvb_inode;
1481 LDLM_DEBUG_LIMIT(inode->i_state & I_FREEING ? D_INFO :
1482 D_WARNING, lock, "lr_lvb_inode %p is "
1483 "bogus: magic %08x",
1484 lock->l_resource->lr_lvb_inode,
1485 lli->lli_inode_magic);
1489 unlock_res_and_lock(lock);
1493 void ll_dir_clear_lsm_md(struct inode *inode)
1495 struct ll_inode_info *lli = ll_i2info(inode);
1497 LASSERT(S_ISDIR(inode->i_mode));
1499 if (lli->lli_lsm_md) {
1500 lmv_free_memmd(lli->lli_lsm_md);
1501 lli->lli_lsm_md = NULL;
1504 if (lli->lli_default_lsm_md) {
1505 lmv_free_memmd(lli->lli_default_lsm_md);
1506 lli->lli_default_lsm_md = NULL;
1510 static struct inode *ll_iget_anon_dir(struct super_block *sb,
1511 const struct lu_fid *fid,
1512 struct lustre_md *md)
1514 struct ll_sb_info *sbi = ll_s2sbi(sb);
1515 struct ll_inode_info *lli;
1516 struct mdt_body *body = md->body;
1517 struct inode *inode;
1523 ino = cl_fid_build_ino(fid, test_bit(LL_SBI_32BIT_API, sbi->ll_flags));
1524 inode = iget_locked(sb, ino);
1525 if (inode == NULL) {
1526 CERROR("%s: failed get simple inode "DFID": rc = -ENOENT\n",
1527 sbi->ll_fsname, PFID(fid));
1528 RETURN(ERR_PTR(-ENOENT));
1531 lli = ll_i2info(inode);
1532 if (inode->i_state & I_NEW) {
1533 inode->i_mode = (inode->i_mode & ~S_IFMT) |
1534 (body->mbo_mode & S_IFMT);
1535 LASSERTF(S_ISDIR(inode->i_mode), "Not slave inode "DFID"\n",
1538 inode->i_mtime.tv_sec = 0;
1539 inode->i_atime.tv_sec = 0;
1540 inode->i_ctime.tv_sec = 0;
1543 #ifdef HAVE_BACKING_DEV_INFO
1544 /* initializing backing dev info. */
1545 inode->i_mapping->backing_dev_info =
1546 &s2lsi(inode->i_sb)->lsi_bdi;
1548 inode->i_op = &ll_dir_inode_operations;
1549 inode->i_fop = &ll_dir_operations;
1550 lli->lli_fid = *fid;
1553 /* master object FID */
1554 lli->lli_pfid = body->mbo_fid1;
1555 CDEBUG(D_INODE, "lli %p slave "DFID" master "DFID"\n",
1556 lli, PFID(fid), PFID(&lli->lli_pfid));
1557 unlock_new_inode(inode);
1559 /* in directory restripe/auto-split, a directory will be
1560 * transformed to a stripe if it's plain, set its pfid here,
1561 * otherwise ll_lock_cancel_bits() can't find the master inode.
1563 lli->lli_pfid = body->mbo_fid1;
1569 static int ll_init_lsm_md(struct inode *inode, struct lustre_md *md)
1572 struct lmv_stripe_md *lsm = md->lmv;
1573 struct ll_inode_info *lli = ll_i2info(inode);
1576 LASSERT(lsm != NULL);
1578 CDEBUG(D_INODE, "%s: "DFID" set dir layout:\n",
1579 ll_i2sbi(inode)->ll_fsname, PFID(&lli->lli_fid));
1580 lsm_md_dump(D_INODE, lsm);
1582 if (!lmv_dir_striped(lsm))
1585 /* XXX sigh, this lsm_root initialization should be in
1586 * LMV layer, but it needs ll_iget right now, so we
1587 * put this here right now. */
1588 for (i = 0; i < lsm->lsm_md_stripe_count; i++) {
1589 fid = &lsm->lsm_md_oinfo[i].lmo_fid;
1590 LASSERT(lsm->lsm_md_oinfo[i].lmo_root == NULL);
1592 if (!fid_is_sane(fid))
1595 /* Unfortunately ll_iget will call ll_update_inode,
1596 * where the initialization of slave inode is slightly
1597 * different, so it reset lsm_md to NULL to avoid
1598 * initializing lsm for slave inode. */
1599 lsm->lsm_md_oinfo[i].lmo_root =
1600 ll_iget_anon_dir(inode->i_sb, fid, md);
1601 if (IS_ERR(lsm->lsm_md_oinfo[i].lmo_root)) {
1602 int rc = PTR_ERR(lsm->lsm_md_oinfo[i].lmo_root);
1604 lsm->lsm_md_oinfo[i].lmo_root = NULL;
1606 iput(lsm->lsm_md_oinfo[i].lmo_root);
1607 lsm->lsm_md_oinfo[i].lmo_root = NULL;
1613 lli->lli_lsm_md = lsm;
1618 static void ll_update_default_lsm_md(struct inode *inode, struct lustre_md *md)
1620 struct ll_inode_info *lli = ll_i2info(inode);
1624 if (!md->default_lmv) {
1625 /* clear default lsm */
1626 if (lli->lli_default_lsm_md) {
1627 down_write(&lli->lli_lsm_sem);
1628 if (lli->lli_default_lsm_md) {
1629 lmv_free_memmd(lli->lli_default_lsm_md);
1630 lli->lli_default_lsm_md = NULL;
1632 up_write(&lli->lli_lsm_sem);
1637 if (lli->lli_default_lsm_md) {
1638 /* do nonthing if default lsm isn't changed */
1639 down_read(&lli->lli_lsm_sem);
1640 if (lli->lli_default_lsm_md &&
1641 lsm_md_eq(lli->lli_default_lsm_md, md->default_lmv)) {
1642 up_read(&lli->lli_lsm_sem);
1645 up_read(&lli->lli_lsm_sem);
1648 down_write(&lli->lli_lsm_sem);
1649 if (lli->lli_default_lsm_md)
1650 lmv_free_memmd(lli->lli_default_lsm_md);
1651 lli->lli_default_lsm_md = md->default_lmv;
1652 lsm_md_dump(D_INODE, md->default_lmv);
1653 md->default_lmv = NULL;
1654 up_write(&lli->lli_lsm_sem);
1658 static int ll_update_lsm_md(struct inode *inode, struct lustre_md *md)
1660 struct ll_inode_info *lli = ll_i2info(inode);
1661 struct lmv_stripe_md *lsm = md->lmv;
1662 struct cl_attr *attr;
1667 LASSERT(S_ISDIR(inode->i_mode));
1668 CDEBUG(D_INODE, "update lsm %p of "DFID"\n", lli->lli_lsm_md,
1669 PFID(ll_inode2fid(inode)));
1671 /* update default LMV */
1672 if (md->default_lmv)
1673 ll_update_default_lsm_md(inode, md);
1675 /* after dir migration/restripe, a stripe may be turned into a
1676 * directory, in this case, zero out its lli_pfid.
1678 if (unlikely(fid_is_norm(&lli->lli_pfid)))
1679 fid_zero(&lli->lli_pfid);
1682 * no striped information from request, lustre_md from req does not
1683 * include stripeEA, see ll_md_setattr()
1689 * normally dir layout doesn't change, only take read lock to check
1690 * that to avoid blocking other MD operations.
1692 down_read(&lli->lli_lsm_sem);
1694 /* some current lookup initialized lsm, and unchanged */
1695 if (lli->lli_lsm_md && lsm_md_eq(lli->lli_lsm_md, lsm))
1696 GOTO(unlock, rc = 0);
1698 /* if dir layout doesn't match, check whether version is increased,
1699 * which means layout is changed, this happens in dir split/merge and
1702 * foreign LMV should not change.
1704 if (lli->lli_lsm_md && lmv_dir_striped(lli->lli_lsm_md) &&
1705 lsm->lsm_md_layout_version <=
1706 lli->lli_lsm_md->lsm_md_layout_version) {
1707 CERROR("%s: "DFID" dir layout mismatch:\n",
1708 ll_i2sbi(inode)->ll_fsname, PFID(&lli->lli_fid));
1709 lsm_md_dump(D_ERROR, lli->lli_lsm_md);
1710 lsm_md_dump(D_ERROR, lsm);
1711 GOTO(unlock, rc = -EINVAL);
1714 up_read(&lli->lli_lsm_sem);
1715 down_write(&lli->lli_lsm_sem);
1716 /* clear existing lsm */
1717 if (lli->lli_lsm_md) {
1718 lmv_free_memmd(lli->lli_lsm_md);
1719 lli->lli_lsm_md = NULL;
1722 rc = ll_init_lsm_md(inode, md);
1723 up_write(&lli->lli_lsm_sem);
1728 /* set md->lmv to NULL, so the following free lustre_md will not free
1733 /* md_merge_attr() may take long, since lsm is already set, switch to
1736 down_read(&lli->lli_lsm_sem);
1738 if (!lmv_dir_striped(lli->lli_lsm_md))
1739 GOTO(unlock, rc = 0);
1741 OBD_ALLOC_PTR(attr);
1743 GOTO(unlock, rc = -ENOMEM);
1745 /* validate the lsm */
1746 rc = md_merge_attr(ll_i2mdexp(inode), lli->lli_lsm_md, attr,
1747 ll_md_blocking_ast);
1749 if (md->body->mbo_valid & OBD_MD_FLNLINK)
1750 md->body->mbo_nlink = attr->cat_nlink;
1751 if (md->body->mbo_valid & OBD_MD_FLSIZE)
1752 md->body->mbo_size = attr->cat_size;
1753 if (md->body->mbo_valid & OBD_MD_FLATIME)
1754 md->body->mbo_atime = attr->cat_atime;
1755 if (md->body->mbo_valid & OBD_MD_FLCTIME)
1756 md->body->mbo_ctime = attr->cat_ctime;
1757 if (md->body->mbo_valid & OBD_MD_FLMTIME)
1758 md->body->mbo_mtime = attr->cat_mtime;
1764 up_read(&lli->lli_lsm_sem);
1769 void ll_clear_inode(struct inode *inode)
1771 struct ll_inode_info *lli = ll_i2info(inode);
1772 struct ll_sb_info *sbi = ll_i2sbi(inode);
1776 CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p)\n",
1777 PFID(ll_inode2fid(inode)), inode);
1779 if (S_ISDIR(inode->i_mode)) {
1780 /* these should have been cleared in ll_file_release */
1781 LASSERT(lli->lli_opendir_key == NULL);
1782 LASSERT(lli->lli_sai == NULL);
1783 LASSERT(lli->lli_opendir_pid == 0);
1785 pcc_inode_free(inode);
1788 md_null_inode(sbi->ll_md_exp, ll_inode2fid(inode));
1790 LASSERT(!lli->lli_open_fd_write_count);
1791 LASSERT(!lli->lli_open_fd_read_count);
1792 LASSERT(!lli->lli_open_fd_exec_count);
1794 if (lli->lli_mds_write_och)
1795 ll_md_real_close(inode, FMODE_WRITE);
1796 if (lli->lli_mds_exec_och)
1797 ll_md_real_close(inode, FMODE_EXEC);
1798 if (lli->lli_mds_read_och)
1799 ll_md_real_close(inode, FMODE_READ);
1801 if (S_ISLNK(inode->i_mode) && lli->lli_symlink_name) {
1802 OBD_FREE(lli->lli_symlink_name,
1803 strlen(lli->lli_symlink_name) + 1);
1804 lli->lli_symlink_name = NULL;
1807 ll_xattr_cache_destroy(inode);
1809 forget_all_cached_acls(inode);
1811 lli->lli_inode_magic = LLI_INODE_DEAD;
1813 if (S_ISDIR(inode->i_mode))
1814 ll_dir_clear_lsm_md(inode);
1815 else if (S_ISREG(inode->i_mode) && !is_bad_inode(inode))
1816 LASSERT(list_empty(&lli->lli_agl_list));
1819 * XXX This has to be done before lsm is freed below, because
1820 * cl_object still uses inode lsm.
1822 cl_inode_fini(inode);
1824 llcrypt_put_encryption_info(inode);
1829 static int ll_md_setattr(struct dentry *dentry, struct md_op_data *op_data)
1831 struct lustre_md md;
1832 struct inode *inode = dentry->d_inode;
1833 struct ll_sb_info *sbi = ll_i2sbi(inode);
1834 struct ptlrpc_request *request = NULL;
1839 op_data = ll_prep_md_op_data(op_data, inode, NULL, NULL, 0, 0,
1840 LUSTRE_OPC_ANY, NULL);
1841 if (IS_ERR(op_data))
1842 RETURN(PTR_ERR(op_data));
1844 /* If this is a chgrp of a regular file, we want to reserve enough
1845 * quota to cover the entire file size.
1847 if (S_ISREG(inode->i_mode) && op_data->op_attr.ia_valid & ATTR_GID &&
1848 from_kgid(&init_user_ns, op_data->op_attr.ia_gid) !=
1849 from_kgid(&init_user_ns, inode->i_gid)) {
1850 op_data->op_xvalid |= OP_XVALID_BLOCKS;
1851 op_data->op_attr_blocks = inode->i_blocks;
1855 rc = md_setattr(sbi->ll_md_exp, op_data, NULL, 0, &request);
1857 ptlrpc_req_finished(request);
1858 if (rc == -ENOENT) {
1860 /* Unlinked special device node? Or just a race?
1861 * Pretend we done everything. */
1862 if (!S_ISREG(inode->i_mode) &&
1863 !S_ISDIR(inode->i_mode)) {
1864 ia_valid = op_data->op_attr.ia_valid;
1865 op_data->op_attr.ia_valid &= ~TIMES_SET_FLAGS;
1866 rc = simple_setattr(&init_user_ns, dentry,
1868 op_data->op_attr.ia_valid = ia_valid;
1870 } else if (rc != -EPERM && rc != -EACCES && rc != -ETXTBSY) {
1871 CERROR("md_setattr fails: rc = %d\n", rc);
1876 rc = md_get_lustre_md(sbi->ll_md_exp, &request->rq_pill, sbi->ll_dt_exp,
1877 sbi->ll_md_exp, &md);
1879 ptlrpc_req_finished(request);
1883 ia_valid = op_data->op_attr.ia_valid;
1884 /* inode size will be in ll_setattr_ost, can't do it now since dirty
1885 * cache is not cleared yet. */
1886 op_data->op_attr.ia_valid &= ~(TIMES_SET_FLAGS | ATTR_SIZE);
1887 if (S_ISREG(inode->i_mode))
1889 rc = simple_setattr(&init_user_ns, dentry, &op_data->op_attr);
1890 if (S_ISREG(inode->i_mode))
1891 inode_unlock(inode);
1892 op_data->op_attr.ia_valid = ia_valid;
1894 rc = ll_update_inode(inode, &md);
1895 ptlrpc_req_finished(request);
1901 * Zero portion of page that is part of @inode.
1902 * This implies, if necessary:
1903 * - taking cl_lock on range corresponding to concerned page
1904 * - grabbing vm page
1905 * - associating cl_page
1906 * - proceeding to clio read
1907 * - zeroing range in page
1908 * - proceeding to cl_page flush
1909 * - releasing cl_lock
1911 * \param[in] inode inode
1912 * \param[in] index page index
1913 * \param[in] offset offset in page to start zero from
1914 * \param[in] len len to zero
1916 * \retval 0 on success
1917 * \retval negative errno on failure
1919 int ll_io_zero_page(struct inode *inode, pgoff_t index, pgoff_t offset,
1922 struct ll_inode_info *lli = ll_i2info(inode);
1923 struct cl_object *clob = lli->lli_clob;
1925 struct lu_env *env = NULL;
1926 struct cl_io *io = NULL;
1927 struct cl_page *clpage = NULL;
1928 struct page *vmpage = NULL;
1929 unsigned from = index << PAGE_SHIFT;
1930 struct cl_lock *lock = NULL;
1931 struct cl_lock_descr *descr = NULL;
1932 struct cl_2queue *queue = NULL;
1933 struct cl_sync_io *anchor = NULL;
1934 bool holdinglock = false;
1939 env = cl_env_get(&refcheck);
1941 RETURN(PTR_ERR(env));
1943 io = vvp_env_thread_io(env);
1945 rc = cl_io_rw_init(env, io, CIT_WRITE, from, PAGE_SIZE);
1949 lock = vvp_env_lock(env);
1950 descr = &lock->cll_descr;
1951 descr->cld_obj = io->ci_obj;
1952 descr->cld_start = cl_index(io->ci_obj, from);
1953 descr->cld_end = cl_index(io->ci_obj, from + PAGE_SIZE - 1);
1954 descr->cld_mode = CLM_WRITE;
1955 descr->cld_enq_flags = CEF_MUST | CEF_NONBLOCK;
1957 /* request lock for page */
1958 rc = cl_lock_request(env, io, lock);
1959 /* -ECANCELED indicates a matching lock with a different extent
1960 * was already present, and -EEXIST indicates a matching lock
1961 * on exactly the same extent was already present.
1962 * In both cases it means we are covered.
1964 if (rc == -ECANCELED || rc == -EEXIST)
1972 vmpage = grab_cache_page_nowait(inode->i_mapping, index);
1974 GOTO(rellock, rc = -EOPNOTSUPP);
1976 if (!PageDirty(vmpage)) {
1977 /* associate cl_page */
1978 clpage = cl_page_find(env, clob, vmpage->index,
1979 vmpage, CPT_CACHEABLE);
1981 GOTO(pagefini, rc = PTR_ERR(clpage));
1983 cl_page_assume(env, io, clpage);
1986 if (!PageUptodate(vmpage) && !PageDirty(vmpage) &&
1987 !PageWriteback(vmpage)) {
1989 /* Set PagePrivate2 to detect special case of empty page
1990 * in osc_brw_fini_request().
1991 * It is also used to tell ll_io_read_page() that we do not
1992 * want the vmpage to be unlocked.
1994 SetPagePrivate2(vmpage);
1995 rc = ll_io_read_page(env, io, clpage, NULL);
1996 if (!PagePrivate2(vmpage)) {
1997 /* PagePrivate2 was cleared in osc_brw_fini_request()
1998 * meaning we read an empty page. In this case, in order
1999 * to avoid allocating unnecessary block in truncated
2000 * file, we must not zero and write as below. Subsequent
2001 * server-side truncate will handle things correctly.
2003 cl_page_unassume(env, io, clpage);
2004 GOTO(clpfini, rc = 0);
2006 ClearPagePrivate2(vmpage);
2011 /* Thanks to PagePrivate2 flag, ll_io_read_page() did not unlock
2012 * the vmpage, so we are good to proceed and zero range in page.
2014 zero_user(vmpage, offset, len);
2016 if (holdinglock && clpage) {
2017 /* explicitly write newly modified page */
2018 queue = &io->ci_queue;
2019 cl_2queue_init(queue);
2020 anchor = &vvp_env_info(env)->vti_anchor;
2021 cl_sync_io_init(anchor, 1);
2022 clpage->cp_sync_io = anchor;
2023 cl_2queue_add(queue, clpage, true);
2024 rc = cl_io_submit_rw(env, io, CRT_WRITE, queue);
2026 GOTO(queuefini1, rc);
2027 rc = cl_sync_io_wait(env, anchor, 0);
2029 GOTO(queuefini2, rc);
2030 cl_page_assume(env, io, clpage);
2033 cl_2queue_discard(env, io, queue);
2035 cl_2queue_disown(env, io, queue);
2036 cl_2queue_fini(env, queue);
2041 cl_page_put(env, clpage);
2043 unlock_page(vmpage);
2047 cl_lock_release(env, lock);
2049 cl_io_fini(env, io);
2052 cl_env_put(env, &refcheck);
2058 * Get reference file from volatile file name.
2059 * Volatile file name may look like:
2060 * <parent>/LUSTRE_VOLATILE_HDR:<mdt_index>:<random>:fd=<fd>
2061 * where fd is opened descriptor of reference file.
2063 * \param[in] volatile_name volatile file name
2064 * \param[in] volatile_len volatile file name length
2065 * \param[out] ref_file pointer to struct file of reference file
2067 * \retval 0 on success
2068 * \retval negative errno on failure
2070 int volatile_ref_file(const char *volatile_name, int volatile_len,
2071 struct file **ref_file)
2073 char *p, *q, *fd_str;
2076 p = strnstr(volatile_name, ":fd=", volatile_len);
2077 if (!p || strlen(p + 4) == 0)
2080 q = strchrnul(p + 4, ':');
2081 fd_str = kstrndup(p + 4, q - p - 4, GFP_NOFS);
2084 rc = kstrtouint(fd_str, 10, &fd);
2089 *ref_file = fget(fd);
2095 /* If this inode has objects allocated to it (lsm != NULL), then the OST
2096 * object(s) determine the file size and mtime. Otherwise, the MDS will
2097 * keep these values until such a time that objects are allocated for it.
2098 * We do the MDS operations first, as it is checking permissions for us.
2099 * We don't to the MDS RPC if there is nothing that we want to store there,
2100 * otherwise there is no harm in updating mtime/atime on the MDS if we are
2101 * going to do an RPC anyways.
2103 * If we are doing a truncate, we will send the mtime and ctime updates
2104 * to the OST with the punch RPC, otherwise we do an explicit setattr RPC.
2105 * I don't believe it is possible to get e.g. ATTR_MTIME_SET and ATTR_SIZE
2108 * In case of HSMimport, we only set attr on MDS.
2110 int ll_setattr_raw(struct dentry *dentry, struct iattr *attr,
2111 enum op_xvalid xvalid, bool hsm_import)
2113 struct inode *inode = dentry->d_inode;
2114 struct ll_inode_info *lli = ll_i2info(inode);
2115 struct md_op_data *op_data = NULL;
2116 ktime_t kstart = ktime_get();
2121 CDEBUG(D_VFSTRACE, "%s: setattr inode "DFID"(%p) from %llu to %llu, "
2122 "valid %x, hsm_import %d\n",
2123 ll_i2sbi(inode)->ll_fsname, PFID(&lli->lli_fid),
2124 inode, i_size_read(inode), attr->ia_size, attr->ia_valid,
2127 if (attr->ia_valid & ATTR_SIZE) {
2128 /* Check new size against VFS/VM file size limit and rlimit */
2129 rc = inode_newsize_ok(inode, attr->ia_size);
2133 /* The maximum Lustre file size is variable, based on the
2134 * OST maximum object size and number of stripes. This
2135 * needs another check in addition to the VFS check above. */
2136 if (attr->ia_size > ll_file_maxbytes(inode)) {
2137 CDEBUG(D_INODE,"file "DFID" too large %llu > %llu\n",
2138 PFID(&lli->lli_fid), attr->ia_size,
2139 ll_file_maxbytes(inode));
2143 attr->ia_valid |= ATTR_MTIME | ATTR_CTIME;
2146 /* POSIX: check before ATTR_*TIME_SET set (from inode_change_ok) */
2147 if (attr->ia_valid & TIMES_SET_FLAGS) {
2148 if ((!uid_eq(current_fsuid(), inode->i_uid)) &&
2149 !capable(CAP_FOWNER))
2153 /* We mark all of the fields "set" so MDS/OST does not re-set them */
2154 if (!(xvalid & OP_XVALID_CTIME_SET) &&
2155 (attr->ia_valid & ATTR_CTIME)) {
2156 attr->ia_ctime = current_time(inode);
2157 xvalid |= OP_XVALID_CTIME_SET;
2159 if (!(attr->ia_valid & ATTR_ATIME_SET) &&
2160 (attr->ia_valid & ATTR_ATIME)) {
2161 attr->ia_atime = current_time(inode);
2162 attr->ia_valid |= ATTR_ATIME_SET;
2164 if (!(attr->ia_valid & ATTR_MTIME_SET) &&
2165 (attr->ia_valid & ATTR_MTIME)) {
2166 attr->ia_mtime = current_time(inode);
2167 attr->ia_valid |= ATTR_MTIME_SET;
2170 if (attr->ia_valid & (ATTR_MTIME | ATTR_CTIME))
2171 CDEBUG(D_INODE, "setting mtime %lld, ctime %lld, now = %lld\n",
2172 (s64)attr->ia_mtime.tv_sec, (s64)attr->ia_ctime.tv_sec,
2173 ktime_get_real_seconds());
2175 if (S_ISREG(inode->i_mode))
2176 inode_unlock(inode);
2178 /* We always do an MDS RPC, even if we're only changing the size;
2179 * only the MDS knows whether truncate() should fail with -ETXTBUSY */
2181 OBD_ALLOC_PTR(op_data);
2182 if (op_data == NULL)
2183 GOTO(out, rc = -ENOMEM);
2185 if (!hsm_import && attr->ia_valid & ATTR_SIZE) {
2186 /* If we are changing file size, file content is
2187 * modified, flag it.
2189 xvalid |= OP_XVALID_OWNEROVERRIDE;
2190 op_data->op_bias |= MDS_DATA_MODIFIED;
2191 clear_bit(LLIF_DATA_MODIFIED, &lli->lli_flags);
2194 if (attr->ia_valid & ATTR_FILE) {
2195 struct ll_file_data *fd = attr->ia_file->private_data;
2197 if (fd->fd_lease_och)
2198 op_data->op_bias |= MDS_TRUNC_KEEP_LEASE;
2201 op_data->op_attr = *attr;
2202 op_data->op_xvalid = xvalid;
2204 rc = ll_md_setattr(dentry, op_data);
2208 if (!S_ISREG(inode->i_mode) || hsm_import)
2211 if (attr->ia_valid & (ATTR_SIZE | ATTR_ATIME | ATTR_ATIME_SET |
2212 ATTR_MTIME | ATTR_MTIME_SET | ATTR_CTIME) ||
2213 xvalid & OP_XVALID_CTIME_SET) {
2214 bool cached = false;
2216 rc = pcc_inode_setattr(inode, attr, &cached);
2219 CERROR("%s: PCC inode "DFID" setattr failed: "
2221 ll_i2sbi(inode)->ll_fsname,
2222 PFID(&lli->lli_fid), rc);
2226 unsigned int flags = 0;
2228 /* For truncate and utimes sending attributes to OSTs,
2229 * setting mtime/atime to the past will be performed
2230 * under PW [0:EOF] extent lock (new_size:EOF for
2231 * truncate). It may seem excessive to send mtime/atime
2232 * updates to OSTs when not setting times to past, but
2233 * it is necessary due to possible time
2234 * de-synchronization between MDT inode and OST objects
2236 if (S_ISREG(inode->i_mode) && IS_ENCRYPTED(inode)) {
2237 xvalid |= OP_XVALID_FLAGS;
2238 flags = LUSTRE_ENCRYPT_FL;
2239 /* Call to ll_io_zero_page is not necessary if
2240 * truncating on PAGE_SIZE boundary, because
2241 * whole pages will be wiped.
2242 * In case of Direct IO, all we need is to set
2245 if (attr->ia_valid & ATTR_SIZE &&
2246 attr->ia_size & ~PAGE_MASK &&
2247 !(attr->ia_valid & ATTR_FILE &&
2248 attr->ia_file->f_flags & O_DIRECT)) {
2250 attr->ia_size & (PAGE_SIZE - 1);
2252 rc = ll_io_zero_page(inode,
2253 attr->ia_size >> PAGE_SHIFT,
2254 offset, PAGE_SIZE - offset);
2258 /* If encrypted volatile file without the key,
2259 * we need to fetch size from reference file,
2260 * and set it on OST objects. This happens when
2261 * migrating or extending an encrypted file
2264 if (filename_is_volatile(dentry->d_name.name,
2267 llcrypt_require_key(inode) == -ENOKEY) {
2268 struct file *ref_file;
2269 struct inode *ref_inode;
2270 struct ll_inode_info *ref_lli;
2271 struct cl_object *ref_obj;
2272 struct cl_attr ref_attr = { 0 };
2276 rc = volatile_ref_file(
2277 dentry->d_name.name,
2283 ref_inode = file_inode(ref_file);
2286 GOTO(out, rc = -EINVAL);
2289 env = cl_env_get(&refcheck);
2291 GOTO(out, rc = PTR_ERR(env));
2293 ref_lli = ll_i2info(ref_inode);
2294 ref_obj = ref_lli->lli_clob;
2295 cl_object_attr_lock(ref_obj);
2296 rc = cl_object_attr_get(env, ref_obj,
2298 cl_object_attr_unlock(ref_obj);
2299 cl_env_put(env, &refcheck);
2304 attr->ia_valid |= ATTR_SIZE;
2305 attr->ia_size = ref_attr.cat_size;
2308 rc = cl_setattr_ost(lli->lli_clob, attr, xvalid, flags);
2312 /* If the file was restored, it needs to set dirty flag.
2314 * We've already sent MDS_DATA_MODIFIED flag in
2315 * ll_md_setattr() for truncate. However, the MDT refuses to
2316 * set the HS_DIRTY flag on released files, so we have to set
2317 * it again if the file has been restored. Please check how
2318 * LLIF_DATA_MODIFIED is set in vvp_io_setattr_fini().
2320 * Please notice that if the file is not released, the previous
2321 * MDS_DATA_MODIFIED has taken effect and usually
2322 * LLIF_DATA_MODIFIED is not set(see vvp_io_setattr_fini()).
2323 * This way we can save an RPC for common open + trunc
2325 if (test_and_clear_bit(LLIF_DATA_MODIFIED, &lli->lli_flags)) {
2326 struct hsm_state_set hss = {
2327 .hss_valid = HSS_SETMASK,
2328 .hss_setmask = HS_DIRTY,
2332 rc2 = ll_hsm_state_set(inode, &hss);
2333 /* truncate and write can happen at the same time, so that
2334 * the file can be set modified even though the file is not
2335 * restored from released state, and ll_hsm_state_set() is
2336 * not applicable for the file, and rc2 < 0 is normal in this
2339 CDEBUG(D_INFO, DFID "HSM set dirty failed: rc2 = %d\n",
2340 PFID(ll_inode2fid(inode)), rc2);
2345 if (op_data != NULL)
2346 ll_finish_md_op_data(op_data);
2348 if (S_ISREG(inode->i_mode)) {
2350 if ((attr->ia_valid & ATTR_SIZE) && !hsm_import)
2351 inode_dio_wait(inode);
2352 /* Once we've got the i_mutex, it's safe to set the S_NOSEC
2353 * flag. ll_update_inode (called from ll_md_setattr), clears
2354 * inode flags, so there is a gap where S_NOSEC is not set.
2355 * This can cause a writer to take the i_mutex unnecessarily,
2356 * but this is safe to do and should be rare. */
2357 inode_has_no_xattr(inode);
2361 ll_stats_ops_tally(ll_i2sbi(inode), attr->ia_valid & ATTR_SIZE ?
2362 LPROC_LL_TRUNC : LPROC_LL_SETATTR,
2363 ktime_us_delta(ktime_get(), kstart));
2368 int ll_setattr(struct user_namespace *mnt_userns, struct dentry *de,
2371 int mode = de->d_inode->i_mode;
2372 enum op_xvalid xvalid = 0;
2375 rc = llcrypt_prepare_setattr(de, attr);
2379 if ((attr->ia_valid & (ATTR_CTIME|ATTR_SIZE|ATTR_MODE)) ==
2380 (ATTR_CTIME|ATTR_SIZE|ATTR_MODE))
2381 xvalid |= OP_XVALID_OWNEROVERRIDE;
2383 if (((attr->ia_valid & (ATTR_MODE|ATTR_FORCE|ATTR_SIZE)) ==
2384 (ATTR_SIZE|ATTR_MODE)) &&
2385 (((mode & S_ISUID) && !(attr->ia_mode & S_ISUID)) ||
2386 (((mode & (S_ISGID|S_IXGRP)) == (S_ISGID|S_IXGRP)) &&
2387 !(attr->ia_mode & S_ISGID))))
2388 attr->ia_valid |= ATTR_FORCE;
2390 if ((attr->ia_valid & ATTR_MODE) &&
2392 !(attr->ia_mode & S_ISUID) &&
2393 !(attr->ia_valid & ATTR_KILL_SUID))
2394 attr->ia_valid |= ATTR_KILL_SUID;
2396 if ((attr->ia_valid & ATTR_MODE) &&
2397 ((mode & (S_ISGID|S_IXGRP)) == (S_ISGID|S_IXGRP)) &&
2398 !(attr->ia_mode & S_ISGID) &&
2399 !(attr->ia_valid & ATTR_KILL_SGID))
2400 attr->ia_valid |= ATTR_KILL_SGID;
2402 return ll_setattr_raw(de, attr, xvalid, false);
2405 int ll_statfs_internal(struct ll_sb_info *sbi, struct obd_statfs *osfs,
2408 struct obd_statfs obd_osfs = { 0 };
2413 max_age = ktime_get_seconds() - sbi->ll_statfs_max_age;
2415 if (test_bit(LL_SBI_LAZYSTATFS, sbi->ll_flags))
2416 flags |= OBD_STATFS_NODELAY;
2418 rc = obd_statfs(NULL, sbi->ll_md_exp, osfs, max_age, flags);
2422 osfs->os_type = LL_SUPER_MAGIC;
2424 CDEBUG(D_SUPER, "MDC blocks %llu/%llu objects %llu/%llu\n",
2425 osfs->os_bavail, osfs->os_blocks, osfs->os_ffree, osfs->os_files);
2427 if (osfs->os_state & OS_STATFS_SUM)
2430 rc = obd_statfs(NULL, sbi->ll_dt_exp, &obd_osfs, max_age, flags);
2431 if (rc) /* Possibly a filesystem with no OSTs. Report MDT totals. */
2434 CDEBUG(D_SUPER, "OSC blocks %llu/%llu objects %llu/%llu\n",
2435 obd_osfs.os_bavail, obd_osfs.os_blocks, obd_osfs.os_ffree,
2438 osfs->os_bsize = obd_osfs.os_bsize;
2439 osfs->os_blocks = obd_osfs.os_blocks;
2440 osfs->os_bfree = obd_osfs.os_bfree;
2441 osfs->os_bavail = obd_osfs.os_bavail;
2443 /* If we have _some_ OSTs, but don't have as many free objects on the
2444 * OSTs as inodes on the MDTs, reduce the reported number of inodes
2445 * to compensate, so that the "inodes in use" number is correct.
2446 * This should be kept in sync with lod_statfs() behaviour.
2448 if (obd_osfs.os_files && obd_osfs.os_ffree < osfs->os_ffree) {
2449 osfs->os_files = (osfs->os_files - osfs->os_ffree) +
2451 osfs->os_ffree = obd_osfs.os_ffree;
2458 static int ll_statfs_project(struct inode *inode, struct kstatfs *sfs)
2460 struct if_quotactl qctl = {
2461 .qc_cmd = LUSTRE_Q_GETQUOTA,
2462 .qc_type = PRJQUOTA,
2463 .qc_valid = QC_GENERAL,
2465 u64 limit, curblock;
2468 qctl.qc_id = ll_i2info(inode)->lli_projid;
2469 ret = quotactl_ioctl(inode->i_sb, &qctl);
2471 /* ignore errors if project ID does not have
2472 * a quota limit or feature unsupported.
2474 if (ret == -ESRCH || ret == -EOPNOTSUPP)
2479 limit = ((qctl.qc_dqblk.dqb_bsoftlimit ?
2480 qctl.qc_dqblk.dqb_bsoftlimit :
2481 qctl.qc_dqblk.dqb_bhardlimit) * 1024) / sfs->f_bsize;
2482 if (limit && sfs->f_blocks > limit) {
2483 curblock = (qctl.qc_dqblk.dqb_curspace +
2484 sfs->f_bsize - 1) / sfs->f_bsize;
2485 sfs->f_blocks = limit;
2486 sfs->f_bfree = sfs->f_bavail =
2487 (sfs->f_blocks > curblock) ?
2488 (sfs->f_blocks - curblock) : 0;
2491 limit = qctl.qc_dqblk.dqb_isoftlimit ?
2492 qctl.qc_dqblk.dqb_isoftlimit :
2493 qctl.qc_dqblk.dqb_ihardlimit;
2494 if (limit && sfs->f_files > limit) {
2495 sfs->f_files = limit;
2496 sfs->f_ffree = (sfs->f_files >
2497 qctl.qc_dqblk.dqb_curinodes) ?
2498 (sfs->f_files - qctl.qc_dqblk.dqb_curinodes) : 0;
2504 int ll_statfs(struct dentry *de, struct kstatfs *sfs)
2506 struct super_block *sb = de->d_sb;
2507 struct obd_statfs osfs;
2508 __u64 fsid = huge_encode_dev(sb->s_dev);
2509 ktime_t kstart = ktime_get();
2512 CDEBUG(D_VFSTRACE, "VFS Op:sb=%s (%p)\n", sb->s_id, sb);
2514 /* Some amount of caching on the client is allowed */
2515 rc = ll_statfs_internal(ll_s2sbi(sb), &osfs, OBD_STATFS_SUM);
2519 statfs_unpack(sfs, &osfs);
2521 /* We need to downshift for all 32-bit kernels, because we can't
2522 * tell if the kernel is being called via sys_statfs64() or not.
2523 * Stop before overflowing f_bsize - in which case it is better
2524 * to just risk EOVERFLOW if caller is using old sys_statfs(). */
2525 if (sizeof(long) < 8) {
2526 while (osfs.os_blocks > ~0UL && sfs->f_bsize < 0x40000000) {
2529 osfs.os_blocks >>= 1;
2530 osfs.os_bfree >>= 1;
2531 osfs.os_bavail >>= 1;
2535 sfs->f_blocks = osfs.os_blocks;
2536 sfs->f_bfree = osfs.os_bfree;
2537 sfs->f_bavail = osfs.os_bavail;
2538 sfs->f_fsid.val[0] = (__u32)fsid;
2539 sfs->f_fsid.val[1] = (__u32)(fsid >> 32);
2540 if (ll_i2info(de->d_inode)->lli_projid)
2541 return ll_statfs_project(de->d_inode, sfs);
2543 ll_stats_ops_tally(ll_s2sbi(sb), LPROC_LL_STATFS,
2544 ktime_us_delta(ktime_get(), kstart));
2549 void ll_inode_size_lock(struct inode *inode)
2551 struct ll_inode_info *lli;
2553 LASSERT(!S_ISDIR(inode->i_mode));
2555 lli = ll_i2info(inode);
2556 mutex_lock(&lli->lli_size_mutex);
2559 void ll_inode_size_unlock(struct inode *inode)
2561 struct ll_inode_info *lli;
2563 lli = ll_i2info(inode);
2564 mutex_unlock(&lli->lli_size_mutex);
2567 void ll_update_inode_flags(struct inode *inode, unsigned int ext_flags)
2569 /* do not clear encryption flag */
2570 ext_flags |= ll_inode_to_ext_flags(inode->i_flags) & LUSTRE_ENCRYPT_FL;
2571 inode->i_flags = ll_ext_to_inode_flags(ext_flags);
2572 if (ext_flags & LUSTRE_PROJINHERIT_FL)
2573 set_bit(LLIF_PROJECT_INHERIT, &ll_i2info(inode)->lli_flags);
2575 clear_bit(LLIF_PROJECT_INHERIT, &ll_i2info(inode)->lli_flags);
2578 int ll_update_inode(struct inode *inode, struct lustre_md *md)
2580 struct ll_inode_info *lli = ll_i2info(inode);
2581 struct mdt_body *body = md->body;
2582 struct ll_sb_info *sbi = ll_i2sbi(inode);
2586 if (body->mbo_valid & OBD_MD_FLEASIZE) {
2587 rc = cl_file_inode_init(inode, md);
2592 if (S_ISDIR(inode->i_mode)) {
2593 rc = ll_update_lsm_md(inode, md);
2598 if (body->mbo_valid & OBD_MD_FLACL)
2599 lli_replace_acl(lli, md);
2601 api32 = test_bit(LL_SBI_32BIT_API, sbi->ll_flags);
2602 inode->i_ino = cl_fid_build_ino(&body->mbo_fid1, api32);
2603 inode->i_generation = cl_fid_build_gen(&body->mbo_fid1);
2605 if (body->mbo_valid & OBD_MD_FLATIME) {
2606 if (body->mbo_atime > inode->i_atime.tv_sec)
2607 inode->i_atime.tv_sec = body->mbo_atime;
2608 lli->lli_atime = body->mbo_atime;
2611 if (body->mbo_valid & OBD_MD_FLMTIME) {
2612 if (body->mbo_mtime > inode->i_mtime.tv_sec) {
2614 "setting ino %lu mtime from %lld to %llu\n",
2615 inode->i_ino, (s64)inode->i_mtime.tv_sec,
2617 inode->i_mtime.tv_sec = body->mbo_mtime;
2619 lli->lli_mtime = body->mbo_mtime;
2622 if (body->mbo_valid & OBD_MD_FLCTIME) {
2623 if (body->mbo_ctime > inode->i_ctime.tv_sec)
2624 inode->i_ctime.tv_sec = body->mbo_ctime;
2625 lli->lli_ctime = body->mbo_ctime;
2628 if (body->mbo_valid & OBD_MD_FLBTIME)
2629 lli->lli_btime = body->mbo_btime;
2631 /* Clear i_flags to remove S_NOSEC before permissions are updated */
2632 if (body->mbo_valid & OBD_MD_FLFLAGS)
2633 ll_update_inode_flags(inode, body->mbo_flags);
2634 if (body->mbo_valid & OBD_MD_FLMODE)
2635 inode->i_mode = (inode->i_mode & S_IFMT) |
2636 (body->mbo_mode & ~S_IFMT);
2638 if (body->mbo_valid & OBD_MD_FLTYPE)
2639 inode->i_mode = (inode->i_mode & ~S_IFMT) |
2640 (body->mbo_mode & S_IFMT);
2642 LASSERT(inode->i_mode != 0);
2643 if (body->mbo_valid & OBD_MD_FLUID)
2644 inode->i_uid = make_kuid(&init_user_ns, body->mbo_uid);
2645 if (body->mbo_valid & OBD_MD_FLGID)
2646 inode->i_gid = make_kgid(&init_user_ns, body->mbo_gid);
2647 if (body->mbo_valid & OBD_MD_FLPROJID)
2648 lli->lli_projid = body->mbo_projid;
2649 if (body->mbo_valid & OBD_MD_FLNLINK) {
2650 spin_lock(&inode->i_lock);
2651 set_nlink(inode, body->mbo_nlink);
2652 spin_unlock(&inode->i_lock);
2654 if (body->mbo_valid & OBD_MD_FLRDEV)
2655 inode->i_rdev = old_decode_dev(body->mbo_rdev);
2657 if (body->mbo_valid & OBD_MD_FLID) {
2658 /* FID shouldn't be changed! */
2659 if (fid_is_sane(&lli->lli_fid)) {
2660 LASSERTF(lu_fid_eq(&lli->lli_fid, &body->mbo_fid1),
2661 "Trying to change FID "DFID
2662 " to the "DFID", inode "DFID"(%p)\n",
2663 PFID(&lli->lli_fid), PFID(&body->mbo_fid1),
2664 PFID(ll_inode2fid(inode)), inode);
2666 lli->lli_fid = body->mbo_fid1;
2670 LASSERT(fid_seq(&lli->lli_fid) != 0);
2672 /* In case of encrypted file without the key, please do not lose
2673 * clear text size stored into lli_lazysize in ll_merge_attr(),
2674 * we will need it in ll_prepare_close().
2676 if (lli->lli_attr_valid & OBD_MD_FLLAZYSIZE && lli->lli_lazysize &&
2677 llcrypt_require_key(inode) == -ENOKEY)
2678 lli->lli_attr_valid = body->mbo_valid | OBD_MD_FLLAZYSIZE;
2680 lli->lli_attr_valid = body->mbo_valid;
2681 if (body->mbo_valid & OBD_MD_FLSIZE) {
2682 i_size_write(inode, body->mbo_size);
2684 CDEBUG(D_VFSTRACE, "inode="DFID", updating i_size %llu\n",
2685 PFID(ll_inode2fid(inode)),
2686 (unsigned long long)body->mbo_size);
2688 if (body->mbo_valid & OBD_MD_FLBLOCKS)
2689 inode->i_blocks = body->mbo_blocks;
2691 if (body->mbo_valid & OBD_MD_FLLAZYSIZE)
2692 lli->lli_lazysize = body->mbo_size;
2693 if (body->mbo_valid & OBD_MD_FLLAZYBLOCKS)
2694 lli->lli_lazyblocks = body->mbo_blocks;
2697 if (body->mbo_valid & OBD_MD_TSTATE) {
2698 /* Set LLIF_FILE_RESTORING if restore ongoing and
2699 * clear it when done to ensure to start again
2700 * glimpsing updated attrs
2702 if (body->mbo_t_state & MS_RESTORE)
2703 set_bit(LLIF_FILE_RESTORING, &lli->lli_flags);
2705 clear_bit(LLIF_FILE_RESTORING, &lli->lli_flags);
2711 /* update directory depth to ROOT, called after LOOKUP lock is fetched. */
2712 void ll_update_dir_depth(struct inode *dir, struct inode *inode)
2714 struct ll_inode_info *lli;
2716 if (!S_ISDIR(inode->i_mode))
2722 lli = ll_i2info(inode);
2723 lli->lli_dir_depth = ll_i2info(dir)->lli_dir_depth + 1;
2724 CDEBUG(D_INODE, DFID" depth %hu\n",
2725 PFID(&lli->lli_fid), lli->lli_dir_depth);
2728 void ll_truncate_inode_pages_final(struct inode *inode)
2730 struct address_space *mapping = &inode->i_data;
2731 unsigned long nrpages;
2732 unsigned long flags;
2734 truncate_inode_pages_final(mapping);
2736 /* Workaround for LU-118: Note nrpages may not be totally updated when
2737 * truncate_inode_pages() returns, as there can be a page in the process
2738 * of deletion (inside __delete_from_page_cache()) in the specified
2739 * range. Thus mapping->nrpages can be non-zero when this function
2740 * returns even after truncation of the whole mapping. Only do this if
2741 * npages isn't already zero.
2743 nrpages = mapping->nrpages;
2745 ll_xa_lock_irqsave(&mapping->i_pages, flags);
2746 nrpages = mapping->nrpages;
2747 ll_xa_unlock_irqrestore(&mapping->i_pages, flags);
2748 } /* Workaround end */
2750 LASSERTF(nrpages == 0, "%s: inode="DFID"(%p) nrpages=%lu, "
2751 "see https://jira.whamcloud.com/browse/LU-118\n",
2752 ll_i2sbi(inode)->ll_fsname,
2753 PFID(ll_inode2fid(inode)), inode, nrpages);
2756 int ll_read_inode2(struct inode *inode, void *opaque)
2758 struct lustre_md *md = opaque;
2759 struct ll_inode_info *lli = ll_i2info(inode);
2763 CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p)\n",
2764 PFID(&lli->lli_fid), inode);
2766 /* Core attributes from the MDS first. This is a new inode, and
2767 * the VFS doesn't zero times in the core inode so we have to do
2768 * it ourselves. They will be overwritten by either MDS or OST
2769 * attributes - we just need to make sure they aren't newer.
2771 inode->i_mtime.tv_sec = 0;
2772 inode->i_atime.tv_sec = 0;
2773 inode->i_ctime.tv_sec = 0;
2775 rc = ll_update_inode(inode, md);
2779 /* OIDEBUG(inode); */
2781 #ifdef HAVE_BACKING_DEV_INFO
2782 /* initializing backing dev info. */
2783 inode->i_mapping->backing_dev_info = &s2lsi(inode->i_sb)->lsi_bdi;
2785 if (S_ISREG(inode->i_mode)) {
2786 struct ll_sb_info *sbi = ll_i2sbi(inode);
2787 inode->i_op = &ll_file_inode_operations;
2788 inode->i_fop = sbi->ll_fop;
2789 inode->i_mapping->a_ops = (struct address_space_operations *)&ll_aops;
2791 } else if (S_ISDIR(inode->i_mode)) {
2792 inode->i_op = &ll_dir_inode_operations;
2793 inode->i_fop = &ll_dir_operations;
2795 } else if (S_ISLNK(inode->i_mode)) {
2796 inode->i_op = &ll_fast_symlink_inode_operations;
2799 inode->i_op = &ll_special_inode_operations;
2801 init_special_inode(inode, inode->i_mode,
2810 void ll_delete_inode(struct inode *inode)
2812 struct ll_inode_info *lli = ll_i2info(inode);
2815 if (S_ISREG(inode->i_mode) && lli->lli_clob != NULL) {
2816 /* It is last chance to write out dirty pages,
2817 * otherwise we may lose data while umount.
2819 * If i_nlink is 0 then just discard data. This is safe because
2820 * local inode gets i_nlink 0 from server only for the last
2821 * unlink, so that file is not opened somewhere else
2823 cl_sync_file_range(inode, 0, OBD_OBJECT_EOF, inode->i_nlink ?
2824 CL_FSYNC_LOCAL : CL_FSYNC_DISCARD, 1);
2827 ll_truncate_inode_pages_final(inode);
2828 ll_clear_inode(inode);
2834 int ll_iocontrol(struct inode *inode, struct file *file,
2835 unsigned int cmd, unsigned long arg)
2837 struct ll_sb_info *sbi = ll_i2sbi(inode);
2838 struct ptlrpc_request *req = NULL;
2843 case FS_IOC_GETFLAGS: {
2844 struct mdt_body *body;
2845 struct md_op_data *op_data;
2847 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL,
2848 0, 0, LUSTRE_OPC_ANY,
2850 if (IS_ERR(op_data))
2851 RETURN(PTR_ERR(op_data));
2853 op_data->op_valid = OBD_MD_FLFLAGS;
2854 rc = md_getattr(sbi->ll_md_exp, op_data, &req);
2855 ll_finish_md_op_data(op_data);
2857 CERROR("%s: failure inode "DFID": rc = %d\n",
2858 sbi->ll_md_exp->exp_obd->obd_name,
2859 PFID(ll_inode2fid(inode)), rc);
2863 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
2865 flags = body->mbo_flags;
2867 ptlrpc_req_finished(req);
2869 RETURN(put_user(flags, (int __user *)arg));
2871 case FS_IOC_SETFLAGS: {
2873 struct md_op_data *op_data;
2874 struct cl_object *obj;
2875 struct fsxattr fa = { 0 };
2877 if (get_user(flags, (int __user *)arg))
2880 fa.fsx_projid = ll_i2info(inode)->lli_projid;
2881 if (flags & LUSTRE_PROJINHERIT_FL)
2882 fa.fsx_xflags = FS_XFLAG_PROJINHERIT;
2884 rc = ll_ioctl_check_project(inode, fa.fsx_xflags,
2889 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
2890 LUSTRE_OPC_ANY, NULL);
2891 if (IS_ERR(op_data))
2892 RETURN(PTR_ERR(op_data));
2894 op_data->op_attr_flags = flags;
2895 op_data->op_xvalid |= OP_XVALID_FLAGS;
2896 rc = md_setattr(sbi->ll_md_exp, op_data, NULL, 0, &req);
2897 ll_finish_md_op_data(op_data);
2898 ptlrpc_req_finished(req);
2902 ll_update_inode_flags(inode, flags);
2904 obj = ll_i2info(inode)->lli_clob;
2908 OBD_ALLOC_PTR(attr);
2912 rc = cl_setattr_ost(obj, attr, OP_XVALID_FLAGS, flags);
2924 int ll_flush_ctx(struct inode *inode)
2926 struct ll_sb_info *sbi = ll_i2sbi(inode);
2928 CDEBUG(D_SEC, "flush context for user %d\n",
2929 from_kuid(&init_user_ns, current_uid()));
2931 obd_set_info_async(NULL, sbi->ll_md_exp,
2932 sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
2934 obd_set_info_async(NULL, sbi->ll_dt_exp,
2935 sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
2940 /* umount -f client means force down, don't save state */
2941 void ll_umount_begin(struct super_block *sb)
2943 struct ll_sb_info *sbi = ll_s2sbi(sb);
2944 struct obd_device *obd;
2945 struct obd_ioctl_data *ioc_data;
2949 CDEBUG(D_VFSTRACE, "VFS Op: superblock %p count %d active %d\n", sb,
2950 sb->s_count, atomic_read(&sb->s_active));
2952 obd = class_exp2obd(sbi->ll_md_exp);
2954 CERROR("Invalid MDC connection handle %#llx\n",
2955 sbi->ll_md_exp->exp_handle.h_cookie);
2961 obd = class_exp2obd(sbi->ll_dt_exp);
2963 CERROR("Invalid LOV connection handle %#llx\n",
2964 sbi->ll_dt_exp->exp_handle.h_cookie);
2970 OBD_ALLOC_PTR(ioc_data);
2972 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_md_exp,
2973 sizeof *ioc_data, ioc_data, NULL);
2975 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_dt_exp,
2976 sizeof *ioc_data, ioc_data, NULL);
2978 OBD_FREE_PTR(ioc_data);
2981 /* Really, we'd like to wait until there are no requests outstanding,
2982 * and then continue. For now, we just periodically checking for vfs
2983 * to decrement mnt_cnt and hope to finish it within 10sec.
2987 !may_umount(sbi->ll_mnt.mnt)) {
2995 int ll_remount_fs(struct super_block *sb, int *flags, char *data)
2997 struct ll_sb_info *sbi = ll_s2sbi(sb);
2998 char *profilenm = get_profile_name(sb);
3002 if ((*flags & MS_RDONLY) != (sb->s_flags & SB_RDONLY)) {
3003 read_only = *flags & MS_RDONLY;
3004 err = obd_set_info_async(NULL, sbi->ll_md_exp,
3005 sizeof(KEY_READ_ONLY),
3006 KEY_READ_ONLY, sizeof(read_only),
3009 LCONSOLE_WARN("Failed to remount %s %s (%d)\n",
3010 profilenm, read_only ?
3011 "read-only" : "read-write", err);
3016 sb->s_flags |= SB_RDONLY;
3018 sb->s_flags &= ~SB_RDONLY;
3020 if (test_bit(LL_SBI_VERBOSE, sbi->ll_flags))
3021 LCONSOLE_WARN("Remounted %s %s\n", profilenm,
3022 read_only ? "read-only" : "read-write");
3028 * Cleanup the open handle that is cached on MDT-side.
3030 * For open case, the client side open handling thread may hit error
3031 * after the MDT grant the open. Under such case, the client should
3032 * send close RPC to the MDT as cleanup; otherwise, the open handle
3033 * on the MDT will be leaked there until the client umount or evicted.
3035 * In further, if someone unlinked the file, because the open handle
3036 * holds the reference on such file/object, then it will block the
3037 * subsequent threads that want to locate such object via FID.
3039 * \param[in] sb super block for this file-system
3040 * \param[in] open_req pointer to the original open request
3042 void ll_open_cleanup(struct super_block *sb, struct req_capsule *pill)
3044 struct mdt_body *body;
3045 struct md_op_data *op_data;
3046 struct ptlrpc_request *close_req = NULL;
3047 struct obd_export *exp = ll_s2sbi(sb)->ll_md_exp;
3050 body = req_capsule_server_get(pill, &RMF_MDT_BODY);
3051 OBD_ALLOC_PTR(op_data);
3052 if (op_data == NULL) {
3053 CWARN("%s: cannot allocate op_data to release open handle for "
3054 DFID"\n", ll_s2sbi(sb)->ll_fsname, PFID(&body->mbo_fid1));
3059 op_data->op_fid1 = body->mbo_fid1;
3060 op_data->op_open_handle = body->mbo_open_handle;
3061 op_data->op_mod_time = ktime_get_real_seconds();
3062 md_close(exp, op_data, NULL, &close_req);
3063 ptlrpc_req_finished(close_req);
3064 ll_finish_md_op_data(op_data);
3069 int ll_prep_inode(struct inode **inode, struct req_capsule *pill,
3070 struct super_block *sb, struct lookup_intent *it)
3072 struct ll_sb_info *sbi = NULL;
3073 struct lustre_md md = { NULL };
3074 bool default_lmv_deleted = false;
3079 LASSERT(*inode || sb);
3080 sbi = sb ? ll_s2sbi(sb) : ll_i2sbi(*inode);
3081 rc = md_get_lustre_md(sbi->ll_md_exp, pill, sbi->ll_dt_exp,
3082 sbi->ll_md_exp, &md);
3087 * clear default_lmv only if intent_getattr reply doesn't contain it.
3088 * but it needs to be done after iget, check this early because
3089 * ll_update_lsm_md() may change md.
3091 if (it && (it->it_op & (IT_LOOKUP | IT_GETATTR)) &&
3092 S_ISDIR(md.body->mbo_mode) && !md.default_lmv)
3093 default_lmv_deleted = true;
3096 rc = ll_update_inode(*inode, &md);
3100 bool api32 = test_bit(LL_SBI_32BIT_API, sbi->ll_flags);
3101 struct lu_fid *fid1 = &md.body->mbo_fid1;
3103 LASSERT(sb != NULL);
3106 * At this point server returns to client's same fid as client
3107 * generated for creating. So using ->fid1 is okay here.
3109 if (!fid_is_sane(fid1)) {
3110 CERROR("%s: Fid is insane "DFID"\n",
3111 sbi->ll_fsname, PFID(fid1));
3112 GOTO(out, rc = -EINVAL);
3115 *inode = ll_iget(sb, cl_fid_build_ino(fid1, api32), &md);
3116 if (IS_ERR(*inode)) {
3118 rc = IS_ERR(*inode) ? PTR_ERR(*inode) : -ENOMEM;
3120 CERROR("new_inode -fatal: rc %d\n", rc);
3125 /* Handling piggyback layout lock.
3126 * Layout lock can be piggybacked by getattr and open request.
3127 * The lsm can be applied to inode only if it comes with a layout lock
3128 * otherwise correct layout may be overwritten, for example:
3129 * 1. proc1: mdt returns a lsm but not granting layout
3130 * 2. layout was changed by another client
3131 * 3. proc2: refresh layout and layout lock granted
3132 * 4. proc1: to apply a stale layout */
3133 if (it != NULL && it->it_lock_mode != 0) {
3134 struct lustre_handle lockh;
3135 struct ldlm_lock *lock;
3137 lockh.cookie = it->it_lock_handle;
3138 lock = ldlm_handle2lock(&lockh);
3139 LASSERT(lock != NULL);
3140 if (ldlm_has_layout(lock)) {
3141 struct cl_object_conf conf;
3143 memset(&conf, 0, sizeof(conf));
3144 conf.coc_opc = OBJECT_CONF_SET;
3145 conf.coc_inode = *inode;
3146 conf.coc_lock = lock;
3147 conf.u.coc_layout = md.layout;
3148 (void)ll_layout_conf(*inode, &conf);
3150 LDLM_LOCK_PUT(lock);
3153 if (default_lmv_deleted)
3154 ll_update_default_lsm_md(*inode, &md);
3156 /* we may want to apply some policy for foreign file/dir */
3157 if (ll_sbi_has_foreign_symlink(sbi)) {
3158 rc = ll_manage_foreign(*inode, &md);
3166 /* cleanup will be done if necessary */
3167 md_free_lustre_md(sbi->ll_md_exp, &md);
3169 if (rc != 0 && it != NULL && it->it_op & IT_OPEN) {
3170 ll_intent_drop_lock(it);
3171 ll_open_cleanup(sb != NULL ? sb : (*inode)->i_sb, pill);
3177 int ll_obd_statfs(struct inode *inode, void __user *arg)
3179 struct ll_sb_info *sbi = NULL;
3180 struct obd_export *exp;
3181 struct obd_ioctl_data *data = NULL;
3186 sbi = ll_i2sbi(inode);
3188 GOTO(out_statfs, rc = -EINVAL);
3190 rc = obd_ioctl_getdata(&data, &len, arg);
3192 GOTO(out_statfs, rc);
3194 if (!data->ioc_inlbuf1 || !data->ioc_inlbuf2 ||
3195 !data->ioc_pbuf1 || !data->ioc_pbuf2)
3196 GOTO(out_statfs, rc = -EINVAL);
3198 if (data->ioc_inllen1 != sizeof(__u32) ||
3199 data->ioc_inllen2 != sizeof(__u32) ||
3200 data->ioc_plen1 != sizeof(struct obd_statfs) ||
3201 data->ioc_plen2 != sizeof(struct obd_uuid))
3202 GOTO(out_statfs, rc = -EINVAL);
3204 memcpy(&type, data->ioc_inlbuf1, sizeof(__u32));
3205 if (type & LL_STATFS_LMV)
3206 exp = sbi->ll_md_exp;
3207 else if (type & LL_STATFS_LOV)
3208 exp = sbi->ll_dt_exp;
3210 GOTO(out_statfs, rc = -ENODEV);
3212 rc = obd_iocontrol(IOC_OBD_STATFS, exp, len, data, NULL);
3214 GOTO(out_statfs, rc);
3216 OBD_FREE_LARGE(data, len);
3221 * this is normally called in ll_fini_md_op_data(), but sometimes it needs to
3222 * be called early to avoid deadlock.
3224 void ll_unlock_md_op_lsm(struct md_op_data *op_data)
3226 if (op_data->op_mea2_sem) {
3227 up_read_non_owner(op_data->op_mea2_sem);
3228 op_data->op_mea2_sem = NULL;
3231 if (op_data->op_mea1_sem) {
3232 up_read_non_owner(op_data->op_mea1_sem);
3233 op_data->op_mea1_sem = NULL;
3237 /* this function prepares md_op_data hint for passing it down to MD stack. */
3238 struct md_op_data *ll_prep_md_op_data(struct md_op_data *op_data,
3239 struct inode *i1, struct inode *i2,
3240 const char *name, size_t namelen,
3241 __u32 mode, enum md_op_code opc,
3244 struct llcrypt_name fname = { 0 };
3247 LASSERT(i1 != NULL);
3250 /* Do not reuse namelen for something else. */
3252 return ERR_PTR(-EINVAL);
3254 if ((!IS_ENCRYPTED(i1) ||
3255 (opc != LUSTRE_OPC_LOOKUP && opc != LUSTRE_OPC_CREATE)) &&
3256 namelen > ll_i2sbi(i1)->ll_namelen)
3257 return ERR_PTR(-ENAMETOOLONG);
3259 /* "/" is not valid name, but it's allowed */
3260 if (!lu_name_is_valid_2(name, namelen) &&
3261 strncmp("/", name, namelen) != 0)
3262 return ERR_PTR(-EINVAL);
3265 if (op_data == NULL)
3266 OBD_ALLOC_PTR(op_data);
3268 if (op_data == NULL)
3269 return ERR_PTR(-ENOMEM);
3271 ll_i2gids(op_data->op_suppgids, i1, i2);
3272 /* If the client is using a subdir mount and looks at what it sees as
3273 * /.fscrypt, interpret it as the .fscrypt dir at the root of the fs.
3275 if (unlikely(i1->i_sb && i1->i_sb->s_root && is_root_inode(i1) &&
3276 !fid_is_root(ll_inode2fid(i1)) &&
3277 name && namelen == strlen(dot_fscrypt_name) &&
3278 strncmp(name, dot_fscrypt_name, namelen) == 0))
3279 lu_root_fid(&op_data->op_fid1);
3281 op_data->op_fid1 = *ll_inode2fid(i1);
3283 if (S_ISDIR(i1->i_mode)) {
3284 down_read_non_owner(&ll_i2info(i1)->lli_lsm_sem);
3285 op_data->op_mea1_sem = &ll_i2info(i1)->lli_lsm_sem;
3286 op_data->op_mea1 = ll_i2info(i1)->lli_lsm_md;
3287 op_data->op_default_mea1 = ll_i2info(i1)->lli_default_lsm_md;
3291 op_data->op_fid2 = *ll_inode2fid(i2);
3292 if (S_ISDIR(i2->i_mode)) {
3294 /* i2 is typically a child of i1, and MUST be
3295 * further from the root to avoid deadlocks.
3297 down_read_non_owner(&ll_i2info(i2)->lli_lsm_sem);
3298 op_data->op_mea2_sem =
3299 &ll_i2info(i2)->lli_lsm_sem;
3301 op_data->op_mea2 = ll_i2info(i2)->lli_lsm_md;
3304 fid_zero(&op_data->op_fid2);
3307 if (test_bit(LL_SBI_64BIT_HASH, ll_i2sbi(i1)->ll_flags))
3308 op_data->op_cli_flags |= CLI_HASH64;
3310 if (ll_need_32bit_api(ll_i2sbi(i1)))
3311 op_data->op_cli_flags |= CLI_API32;
3313 if ((i2 && is_root_inode(i2)) ||
3314 opc == LUSTRE_OPC_LOOKUP || opc == LUSTRE_OPC_CREATE) {
3315 /* In case of lookup, ll_setup_filename() has already been
3316 * called in ll_lookup_it(), so just take provided name.
3317 * Also take provided name if we are dealing with root inode.
3319 fname.disk_name.name = (unsigned char *)name;
3320 fname.disk_name.len = namelen;
3321 } else if (name && namelen) {
3322 struct qstr dname = QSTR_INIT(name, namelen);
3324 struct lu_fid *pfid = NULL;
3328 if (!S_ISDIR(i1->i_mode) && i2 && S_ISDIR(i2->i_mode)) {
3329 /* special case when called from ll_link() */
3334 lookup = (int)(opc == LUSTRE_OPC_ANY);
3336 if (opc == LUSTRE_OPC_ANY && lookup)
3338 rc = ll_setup_filename(dir, &dname, lookup, &fname, pfid);
3340 ll_finish_md_op_data(op_data);
3343 if (pfid && !fid_is_zero(pfid)) {
3345 op_data->op_fid2 = fid;
3346 op_data->op_bias = MDS_FID_OP;
3348 if (fname.disk_name.name &&
3349 fname.disk_name.name != (unsigned char *)name)
3350 /* op_data->op_name must be freed after use */
3351 op_data->op_flags |= MF_OPNAME_KMALLOCED;
3354 /* In fact LUSTRE_OPC_LOOKUP, LUSTRE_OPC_OPEN
3355 * are LUSTRE_OPC_ANY
3357 if (opc == LUSTRE_OPC_LOOKUP || opc == LUSTRE_OPC_OPEN)
3358 op_data->op_code = LUSTRE_OPC_ANY;
3360 op_data->op_code = opc;
3361 op_data->op_name = fname.disk_name.name;
3362 op_data->op_namelen = fname.disk_name.len;
3363 op_data->op_mode = mode;
3364 op_data->op_mod_time = ktime_get_real_seconds();
3365 op_data->op_fsuid = from_kuid(&init_user_ns, current_fsuid());
3366 op_data->op_fsgid = from_kgid(&init_user_ns, current_fsgid());
3367 op_data->op_cap = current_cap();
3368 op_data->op_mds = 0;
3369 if ((opc == LUSTRE_OPC_CREATE) && (name != NULL) &&
3370 filename_is_volatile(name, namelen, &op_data->op_mds)) {
3371 op_data->op_bias |= MDS_CREATE_VOLATILE;
3373 op_data->op_data = data;
3378 void ll_finish_md_op_data(struct md_op_data *op_data)
3380 ll_unlock_md_op_lsm(op_data);
3381 ll_security_release_secctx(op_data->op_file_secctx,
3382 op_data->op_file_secctx_size);
3383 if (op_data->op_flags & MF_OPNAME_KMALLOCED)
3384 /* allocated via ll_setup_filename called
3385 * from ll_prep_md_op_data
3387 kfree(op_data->op_name);
3388 llcrypt_free_ctx(op_data->op_file_encctx, op_data->op_file_encctx_size);
3389 OBD_FREE_PTR(op_data);
3392 int ll_show_options(struct seq_file *seq, struct dentry *dentry)
3394 struct ll_sb_info *sbi;
3397 LASSERT(seq && dentry);
3398 sbi = ll_s2sbi(dentry->d_sb);
3400 if (test_bit(LL_SBI_NOLCK, sbi->ll_flags))
3401 seq_puts(seq, "nolock");
3403 for (i = 1; ll_sbi_flags_name[i].token != LL_SBI_NUM_MOUNT_OPT; i++) {
3404 /* match_table in some cases has patterns for both enabled and
3405 * disabled cases. Ignore 'no'xxx versions if bit is set.
3407 if (test_bit(ll_sbi_flags_name[i].token, sbi->ll_flags) &&
3408 strncmp(ll_sbi_flags_name[i].pattern, "no", 2)) {
3409 if (ll_sbi_flags_name[i].token ==
3410 LL_SBI_FOREIGN_SYMLINK) {
3411 seq_show_option(seq, "foreign_symlink",
3412 sbi->ll_foreign_symlink_prefix);
3414 seq_printf(seq, ",%s",
3415 ll_sbi_flags_name[i].pattern);
3418 /* You can have either localflock or flock but not
3419 * both. If localflock is set don't print flock or
3422 if (ll_sbi_flags_name[i].token == LL_SBI_LOCALFLOCK)
3424 } else if (!test_bit(ll_sbi_flags_name[i].token, sbi->ll_flags) &&
3425 !strncmp(ll_sbi_flags_name[i].pattern, "no", 2)) {
3426 seq_printf(seq, ",%s",
3427 ll_sbi_flags_name[i].pattern);
3435 * Get obd name by cmd, and copy out to user space
3437 int ll_get_obd_name(struct inode *inode, unsigned int cmd, unsigned long arg)
3439 struct ll_sb_info *sbi = ll_i2sbi(inode);
3440 struct obd_device *obd;
3443 if (cmd == OBD_IOC_GETNAME_OLD || cmd == OBD_IOC_GETDTNAME)
3444 obd = class_exp2obd(sbi->ll_dt_exp);
3445 else if (cmd == OBD_IOC_GETMDNAME)
3446 obd = class_exp2obd(sbi->ll_md_exp);
3453 if (copy_to_user((void __user *)arg, obd->obd_name,
3454 strlen(obd->obd_name) + 1))
3461 struct work_struct db_work;
3462 struct dentry *db_dentry;
3463 /* Let's hope the path is not too long, 32 bytes for the work struct
3466 char buf[PAGE_SIZE - sizeof(struct work_struct) - sizeof(void *)];
3469 static void ll_dput_later(struct work_struct *work)
3471 struct dname_buf *db = container_of(work, struct dname_buf, db_work);
3473 dput(db->db_dentry);
3474 free_page((unsigned long)db);
3477 static char* ll_d_path(struct dentry *dentry, char *buf, int bufsize)
3484 p.mnt = current->fs->root.mnt;
3486 path = d_path(&p, buf, bufsize);
3491 void ll_dirty_page_discard_warn(struct inode *inode, int ioret)
3493 struct dname_buf *db;
3495 struct dentry *dentry = NULL;
3497 /* this can be called inside spin lock so use GFP_ATOMIC. */
3498 db = (struct dname_buf *)__get_free_page(GFP_ATOMIC);
3501 dentry = d_find_alias(inode);
3503 path = ll_d_path(dentry, db->buf, sizeof(db->buf));
3506 /* The below message is checked in recovery-small.sh test_24b */
3508 "%s: dirty page discard: %s/fid: "DFID"/%s may get corrupted "
3509 "(rc %d)\n", ll_i2sbi(inode)->ll_fsname,
3510 s2lsi(inode->i_sb)->lsi_lmd->lmd_dev,
3511 PFID(ll_inode2fid(inode)),
3512 (path && !IS_ERR(path)) ? path : "", ioret);
3514 if (dentry != NULL) {
3515 /* We cannot dput here since if we happen to be the last holder
3516 * then we can end up waiting for page evictions that
3517 * in turn wait for RPCs that need this instance of ptlrpcd
3518 * (callng brw_interpret->*page_completion*->vmpage_error->here)
3521 INIT_WORK(&db->db_work, ll_dput_later);
3522 db->db_dentry = dentry;
3523 schedule_work(&db->db_work);
3526 free_page((unsigned long)db);
3530 ssize_t ll_copy_user_md(const struct lov_user_md __user *md,
3531 struct lov_user_md **kbuf)
3533 struct lov_user_md lum;
3537 if (copy_from_user(&lum, md, sizeof(lum)))
3540 lum_size = ll_lov_user_md_size(&lum);
3544 OBD_ALLOC_LARGE(*kbuf, lum_size);
3548 if (copy_from_user(*kbuf, md, lum_size) != 0) {
3549 OBD_FREE_LARGE(*kbuf, lum_size);
3557 * Compute llite root squash state after a change of root squash
3558 * configuration setting or add/remove of a lnet nid
3560 void ll_compute_rootsquash_state(struct ll_sb_info *sbi)
3562 struct root_squash_info *squash = &sbi->ll_squash;
3565 struct lnet_processid id;
3567 /* Update norootsquash flag */
3568 spin_lock(&squash->rsi_lock);
3569 if (list_empty(&squash->rsi_nosquash_nids))
3570 clear_bit(LL_SBI_NOROOTSQUASH, sbi->ll_flags);
3572 /* Do not apply root squash as soon as one of our NIDs is
3573 * in the nosquash_nids list */
3576 while (LNetGetId(i++, &id) != -ENOENT) {
3577 if (nid_is_lo0(&id.nid))
3579 if (cfs_match_nid(lnet_nid_to_nid4(&id.nid),
3580 &squash->rsi_nosquash_nids)) {
3586 set_bit(LL_SBI_NOROOTSQUASH, sbi->ll_flags);
3588 clear_bit(LL_SBI_NOROOTSQUASH, sbi->ll_flags);
3590 spin_unlock(&squash->rsi_lock);
3594 * Parse linkea content to extract information about a given hardlink
3596 * \param[in] ldata - Initialized linkea data
3597 * \param[in] linkno - Link identifier
3598 * \param[out] parent_fid - The entry's parent FID
3599 * \param[out] ln - Entry name destination buffer
3601 * \retval 0 on success
3602 * \retval Appropriate negative error code on failure
3604 static int ll_linkea_decode(struct linkea_data *ldata, unsigned int linkno,
3605 struct lu_fid *parent_fid, struct lu_name *ln)
3611 rc = linkea_init_with_rec(ldata);
3615 if (linkno >= ldata->ld_leh->leh_reccount)
3616 /* beyond last link */
3619 linkea_first_entry(ldata);
3620 for (idx = 0; ldata->ld_lee != NULL; idx++) {
3621 linkea_entry_unpack(ldata->ld_lee, &ldata->ld_reclen, ln,
3626 linkea_next_entry(ldata);
3636 * Get parent FID and name of an identified link. Operation is performed for
3637 * a given link number, letting the caller iterate over linkno to list one or
3638 * all links of an entry.
3640 * \param[in] file - File descriptor against which to perform the operation
3641 * \param[in,out] arg - User-filled structure containing the linkno to operate
3642 * on and the available size. It is eventually filled with
3643 * the requested information or left untouched on error
3645 * \retval - 0 on success
3646 * \retval - Appropriate negative error code on failure
3648 int ll_getparent(struct file *file, struct getparent __user *arg)
3650 struct inode *inode = file_inode(file);
3651 struct linkea_data *ldata;
3652 struct lu_buf buf = LU_BUF_NULL;
3654 struct lu_fid parent_fid;
3661 if (!capable(CAP_DAC_READ_SEARCH) &&
3662 !test_bit(LL_SBI_USER_FID2PATH, ll_i2sbi(inode)->ll_flags))
3665 if (get_user(name_size, &arg->gp_name_size))
3668 if (get_user(linkno, &arg->gp_linkno))
3671 if (name_size > PATH_MAX)
3674 OBD_ALLOC(ldata, sizeof(*ldata));
3678 rc = linkea_data_new(ldata, &buf);
3680 GOTO(ldata_free, rc);
3682 rc = ll_xattr_list(inode, XATTR_NAME_LINK, XATTR_TRUSTED_T, buf.lb_buf,
3683 buf.lb_len, OBD_MD_FLXATTR);
3687 rc = ll_linkea_decode(ldata, linkno, &parent_fid, &ln);
3691 if (ln.ln_namelen >= name_size)
3692 GOTO(lb_free, rc = -EOVERFLOW);
3694 if (copy_to_user(&arg->gp_fid, &parent_fid, sizeof(arg->gp_fid)))
3695 GOTO(lb_free, rc = -EFAULT);
3697 if (copy_to_user(&arg->gp_name, ln.ln_name, ln.ln_namelen))
3698 GOTO(lb_free, rc = -EFAULT);
3700 if (put_user('\0', arg->gp_name + ln.ln_namelen))
3701 GOTO(lb_free, rc = -EFAULT);
3706 OBD_FREE(ldata, sizeof(*ldata));