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/
30 * Lustre is a trademark of Sun Microsystems, Inc.
32 * lustre/llite/llite_lib.c
34 * Lustre Light Super operations
37 #define DEBUG_SUBSYSTEM S_LLITE
39 #include <linux/cpu.h>
40 #include <linux/module.h>
41 #include <linux/random.h>
42 #include <linux/statfs.h>
43 #include <linux/time.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/security.h>
52 #include <linux/fs_struct.h>
54 #ifndef HAVE_CPUS_READ_LOCK
55 #include <libcfs/linux/linux-cpu.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;
102 RETURN(ERR_PTR(-ENOMEM));
104 rc = pcc_super_init(&sbi->ll_pcc_super);
108 spin_lock_init(&sbi->ll_lock);
109 mutex_init(&sbi->ll_lco.lco_lock);
110 spin_lock_init(&sbi->ll_pp_extent_lock);
111 spin_lock_init(&sbi->ll_process_lock);
112 sbi->ll_rw_stats_on = 0;
113 sbi->ll_statfs_max_age = OBD_STATFS_CACHE_SECONDS;
116 pages = si.totalram - si.totalhigh;
117 lru_page_max = pages / 2;
119 sbi->ll_ra_info.ra_async_max_active = ll_get_ra_async_max_active();
120 sbi->ll_ra_info.ll_readahead_wq =
121 cfs_cpt_bind_workqueue("ll-readahead-wq", cfs_cpt_tab,
123 sbi->ll_ra_info.ra_async_max_active);
124 if (IS_ERR(sbi->ll_ra_info.ll_readahead_wq))
125 GOTO(out_pcc, rc = PTR_ERR(sbi->ll_ra_info.ll_readahead_wq));
127 /* initialize ll_cache data */
128 sbi->ll_cache = cl_cache_init(lru_page_max);
129 if (sbi->ll_cache == NULL)
130 GOTO(out_destroy_ra, rc = -ENOMEM);
132 sbi->ll_ra_info.ra_max_pages =
133 min(pages / 32, SBI_DEFAULT_READ_AHEAD_MAX);
134 sbi->ll_ra_info.ra_max_pages_per_file =
135 min(sbi->ll_ra_info.ra_max_pages / 4,
136 SBI_DEFAULT_READ_AHEAD_PER_FILE_MAX);
137 sbi->ll_ra_info.ra_async_pages_per_file_threshold =
138 sbi->ll_ra_info.ra_max_pages_per_file;
139 sbi->ll_ra_info.ra_range_pages = SBI_DEFAULT_RA_RANGE_PAGES;
140 sbi->ll_ra_info.ra_max_read_ahead_whole_pages = -1;
141 atomic_set(&sbi->ll_ra_info.ra_async_inflight, 0);
143 sbi->ll_flags |= LL_SBI_VERBOSE;
144 #ifdef ENABLE_CHECKSUM
145 sbi->ll_flags |= LL_SBI_CHECKSUM;
148 sbi->ll_flags |= LL_SBI_FLOCK;
151 #ifdef HAVE_LRU_RESIZE_SUPPORT
152 sbi->ll_flags |= LL_SBI_LRU_RESIZE;
154 sbi->ll_flags |= LL_SBI_LAZYSTATFS;
156 for (i = 0; i <= LL_PROCESS_HIST_MAX; i++) {
157 spin_lock_init(&sbi->ll_rw_extents_info.pp_extents[i].
159 spin_lock_init(&sbi->ll_rw_extents_info.pp_extents[i].
163 /* metadata statahead is enabled by default */
164 sbi->ll_sa_running_max = LL_SA_RUNNING_DEF;
165 sbi->ll_sa_max = LL_SA_RPC_DEF;
166 atomic_set(&sbi->ll_sa_total, 0);
167 atomic_set(&sbi->ll_sa_wrong, 0);
168 atomic_set(&sbi->ll_sa_running, 0);
169 atomic_set(&sbi->ll_agl_total, 0);
170 sbi->ll_flags |= LL_SBI_AGL_ENABLED;
171 sbi->ll_flags |= LL_SBI_FAST_READ;
172 sbi->ll_flags |= LL_SBI_TINY_WRITE;
173 ll_sbi_set_encrypt(sbi, true);
176 sbi->ll_squash.rsi_uid = 0;
177 sbi->ll_squash.rsi_gid = 0;
178 INIT_LIST_HEAD(&sbi->ll_squash.rsi_nosquash_nids);
179 spin_lock_init(&sbi->ll_squash.rsi_lock);
181 /* Per-filesystem file heat */
182 sbi->ll_heat_decay_weight = SBI_DEFAULT_HEAT_DECAY_WEIGHT;
183 sbi->ll_heat_period_second = SBI_DEFAULT_HEAT_PERIOD_SECOND;
186 destroy_workqueue(sbi->ll_ra_info.ll_readahead_wq);
188 pcc_super_fini(&sbi->ll_pcc_super);
194 static void ll_free_sbi(struct super_block *sb)
196 struct ll_sb_info *sbi = ll_s2sbi(sb);
200 if (!list_empty(&sbi->ll_squash.rsi_nosquash_nids))
201 cfs_free_nidlist(&sbi->ll_squash.rsi_nosquash_nids);
202 if (sbi->ll_ra_info.ll_readahead_wq)
203 destroy_workqueue(sbi->ll_ra_info.ll_readahead_wq);
204 if (sbi->ll_cache != NULL) {
205 cl_cache_decref(sbi->ll_cache);
206 sbi->ll_cache = NULL;
208 pcc_super_fini(&sbi->ll_pcc_super);
209 OBD_FREE(sbi, sizeof(*sbi));
214 static int client_common_fill_super(struct super_block *sb, char *md, char *dt)
216 struct inode *root = NULL;
217 struct ll_sb_info *sbi = ll_s2sbi(sb);
218 struct obd_statfs *osfs = NULL;
219 struct ptlrpc_request *request = NULL;
220 struct obd_connect_data *data = NULL;
221 struct obd_uuid *uuid;
222 struct md_op_data *op_data;
223 struct lustre_md lmd;
225 int size, err, checksum;
228 sbi->ll_md_obd = class_name2obd(md);
229 if (!sbi->ll_md_obd) {
230 CERROR("MD %s: not setup or attached\n", md);
244 /* pass client page size via ocd_grant_blkbits, the server should report
245 * back its backend blocksize for grant calculation purpose */
246 data->ocd_grant_blkbits = PAGE_SHIFT;
248 /* indicate MDT features supported by this client */
249 data->ocd_connect_flags = OBD_CONNECT_IBITS | OBD_CONNECT_NODEVOH |
250 OBD_CONNECT_ATTRFID | OBD_CONNECT_GRANT |
251 OBD_CONNECT_VERSION | OBD_CONNECT_BRW_SIZE |
252 OBD_CONNECT_SRVLOCK | OBD_CONNECT_TRUNCLOCK|
253 OBD_CONNECT_MDS_CAPA | OBD_CONNECT_OSS_CAPA |
254 OBD_CONNECT_CANCELSET | OBD_CONNECT_FID |
255 OBD_CONNECT_AT | OBD_CONNECT_LOV_V3 |
256 OBD_CONNECT_VBR | OBD_CONNECT_FULL20 |
257 OBD_CONNECT_64BITHASH |
258 OBD_CONNECT_EINPROGRESS |
259 OBD_CONNECT_JOBSTATS | OBD_CONNECT_LVB_TYPE |
260 OBD_CONNECT_LAYOUTLOCK | OBD_CONNECT_PINGLESS|
261 OBD_CONNECT_MAX_EASIZE |
262 OBD_CONNECT_FLOCK_DEAD |
263 OBD_CONNECT_DISP_STRIPE | OBD_CONNECT_LFSCK |
264 OBD_CONNECT_OPEN_BY_FID |
265 OBD_CONNECT_DIR_STRIPE |
266 OBD_CONNECT_BULK_MBITS | OBD_CONNECT_CKSUM |
267 OBD_CONNECT_SUBTREE |
268 OBD_CONNECT_MULTIMODRPCS |
269 OBD_CONNECT_GRANT_PARAM |
270 OBD_CONNECT_SHORTIO | OBD_CONNECT_FLAGS2;
272 data->ocd_connect_flags2 = OBD_CONNECT2_DIR_MIGRATE |
273 OBD_CONNECT2_SUM_STATFS |
274 OBD_CONNECT2_OVERSTRIPING |
276 OBD_CONNECT2_LOCK_CONVERT |
277 OBD_CONNECT2_ARCHIVE_ID_ARRAY |
278 OBD_CONNECT2_INC_XID |
280 OBD_CONNECT2_ASYNC_DISCARD |
282 OBD_CONNECT2_CRUSH | OBD_CONNECT2_LSEEK |
283 OBD_CONNECT2_GETATTR_PFID;
285 #ifdef HAVE_LRU_RESIZE_SUPPORT
286 if (sbi->ll_flags & LL_SBI_LRU_RESIZE)
287 data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
289 data->ocd_connect_flags |= OBD_CONNECT_ACL_FLAGS;
291 data->ocd_cksum_types = obd_cksum_types_supported_client();
293 if (OBD_FAIL_CHECK(OBD_FAIL_MDC_LIGHTWEIGHT))
294 /* flag mdc connection as lightweight, only used for test
295 * purpose, use with care */
296 data->ocd_connect_flags |= OBD_CONNECT_LIGHTWEIGHT;
298 data->ocd_ibits_known = MDS_INODELOCK_FULL;
299 data->ocd_version = LUSTRE_VERSION_CODE;
301 if (sb->s_flags & SB_RDONLY)
302 data->ocd_connect_flags |= OBD_CONNECT_RDONLY;
303 if (sbi->ll_flags & LL_SBI_USER_XATTR)
304 data->ocd_connect_flags |= OBD_CONNECT_XATTR;
307 /* Setting this indicates we correctly support S_NOSEC (See kernel
308 * commit 9e1f1de02c2275d7172e18dc4e7c2065777611bf)
310 sb->s_flags |= SB_NOSEC;
313 if (sbi->ll_flags & LL_SBI_FLOCK)
314 sbi->ll_fop = &ll_file_operations_flock;
315 else if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
316 sbi->ll_fop = &ll_file_operations;
318 sbi->ll_fop = &ll_file_operations_noflock;
320 /* always ping even if server suppress_pings */
321 if (sbi->ll_flags & LL_SBI_ALWAYS_PING)
322 data->ocd_connect_flags &= ~OBD_CONNECT_PINGLESS;
324 obd_connect_set_secctx(data);
325 if (ll_sbi_has_encrypt(sbi))
326 obd_connect_set_enc(data);
328 #if defined(CONFIG_SECURITY)
329 data->ocd_connect_flags2 |= OBD_CONNECT2_SELINUX_POLICY;
332 data->ocd_brw_size = MD_MAX_BRW_SIZE;
334 err = obd_connect(NULL, &sbi->ll_md_exp, sbi->ll_md_obd,
335 &sbi->ll_sb_uuid, data, sbi->ll_cache);
337 LCONSOLE_ERROR_MSG(0x14f, "An MDT (md %s) is performing "
338 "recovery, of which this client is not a "
339 "part. Please wait for recovery to complete,"
340 " abort, or time out.\n", md);
343 CERROR("cannot connect to %s: rc = %d\n", md, err);
347 sbi->ll_md_exp->exp_connect_data = *data;
349 err = obd_fid_init(sbi->ll_md_exp->exp_obd, sbi->ll_md_exp,
350 LUSTRE_SEQ_METADATA);
352 CERROR("%s: Can't init metadata layer FID infrastructure, "
353 "rc = %d\n", sbi->ll_md_exp->exp_obd->obd_name, err);
357 /* For mount, we only need fs info from MDT0, and also in DNE, it
358 * can make sure the client can be mounted as long as MDT0 is
360 err = obd_statfs(NULL, sbi->ll_md_exp, osfs,
361 ktime_get_seconds() - sbi->ll_statfs_max_age,
362 OBD_STATFS_FOR_MDT0);
364 GOTO(out_md_fid, err);
366 /* This needs to be after statfs to ensure connect has finished.
367 * Note that "data" does NOT contain the valid connect reply.
368 * If connecting to a 1.8 server there will be no LMV device, so
369 * we can access the MDC export directly and exp_connect_flags will
370 * be non-zero, but if accessing an upgraded 2.1 server it will
371 * have the correct flags filled in.
372 * XXX: fill in the LMV exp_connect_flags from MDC(s). */
373 valid = exp_connect_flags(sbi->ll_md_exp) & CLIENT_CONNECT_MDT_REQD;
374 if (exp_connect_flags(sbi->ll_md_exp) != 0 &&
375 valid != CLIENT_CONNECT_MDT_REQD) {
378 OBD_ALLOC_WAIT(buf, PAGE_SIZE);
379 obd_connect_flags2str(buf, PAGE_SIZE,
380 valid ^ CLIENT_CONNECT_MDT_REQD, 0, ",");
381 LCONSOLE_ERROR_MSG(0x170, "Server %s does not support "
382 "feature(s) needed for correct operation "
383 "of this client (%s). Please upgrade "
384 "server or downgrade client.\n",
385 sbi->ll_md_exp->exp_obd->obd_name, buf);
386 OBD_FREE(buf, PAGE_SIZE);
387 GOTO(out_md_fid, err = -EPROTO);
390 size = sizeof(*data);
391 err = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_CONN_DATA),
392 KEY_CONN_DATA, &size, data);
394 CERROR("%s: Get connect data failed: rc = %d\n",
395 sbi->ll_md_exp->exp_obd->obd_name, err);
396 GOTO(out_md_fid, err);
399 LASSERT(osfs->os_bsize);
400 sb->s_blocksize = osfs->os_bsize;
401 sb->s_blocksize_bits = log2(osfs->os_bsize);
402 sb->s_magic = LL_SUPER_MAGIC;
403 sb->s_maxbytes = MAX_LFS_FILESIZE;
404 sbi->ll_namelen = osfs->os_namelen;
405 sbi->ll_mnt.mnt = current->fs->root.mnt;
407 if ((sbi->ll_flags & LL_SBI_USER_XATTR) &&
408 !(data->ocd_connect_flags & OBD_CONNECT_XATTR)) {
409 LCONSOLE_INFO("Disabling user_xattr feature because "
410 "it is not supported on the server\n");
411 sbi->ll_flags &= ~LL_SBI_USER_XATTR;
414 if (data->ocd_connect_flags & OBD_CONNECT_ACL) {
416 sb->s_flags |= SB_POSIXACL;
418 sbi->ll_flags |= LL_SBI_ACL;
420 LCONSOLE_INFO("client wants to enable acl, but mdt not!\n");
422 sb->s_flags &= ~SB_POSIXACL;
424 sbi->ll_flags &= ~LL_SBI_ACL;
427 if (data->ocd_connect_flags & OBD_CONNECT_64BITHASH)
428 sbi->ll_flags |= LL_SBI_64BIT_HASH;
430 if (data->ocd_connect_flags & OBD_CONNECT_LAYOUTLOCK)
431 sbi->ll_flags |= LL_SBI_LAYOUT_LOCK;
433 if (obd_connect_has_secctx(data))
434 sbi->ll_flags |= LL_SBI_FILE_SECCTX;
436 if (ll_sbi_has_encrypt(sbi) && !obd_connect_has_enc(data)) {
437 if (ll_sbi_has_test_dummy_encryption(sbi))
438 LCONSOLE_WARN("%s: server %s does not support encryption feature, encryption deactivated.\n",
440 sbi->ll_md_exp->exp_obd->obd_name);
441 ll_sbi_set_encrypt(sbi, false);
444 if (data->ocd_ibits_known & MDS_INODELOCK_XATTR) {
445 if (!(data->ocd_connect_flags & OBD_CONNECT_MAX_EASIZE)) {
446 LCONSOLE_INFO("%s: disabling xattr cache due to "
447 "unknown maximum xattr size.\n", dt);
448 } else if (!sbi->ll_xattr_cache_set) {
449 /* If xattr_cache is already set (no matter 0 or 1)
450 * during processing llog, it won't be enabled here. */
451 sbi->ll_flags |= LL_SBI_XATTR_CACHE;
452 sbi->ll_xattr_cache_enabled = 1;
456 sbi->ll_dt_obd = class_name2obd(dt);
457 if (!sbi->ll_dt_obd) {
458 CERROR("DT %s: not setup or attached\n", dt);
459 GOTO(out_md_fid, err = -ENODEV);
462 /* pass client page size via ocd_grant_blkbits, the server should report
463 * back its backend blocksize for grant calculation purpose */
464 data->ocd_grant_blkbits = PAGE_SHIFT;
466 /* indicate OST features supported by this client */
467 data->ocd_connect_flags = OBD_CONNECT_GRANT | OBD_CONNECT_VERSION |
468 OBD_CONNECT_REQPORTAL | OBD_CONNECT_BRW_SIZE |
469 OBD_CONNECT_CANCELSET | OBD_CONNECT_FID |
470 OBD_CONNECT_SRVLOCK | OBD_CONNECT_TRUNCLOCK|
471 OBD_CONNECT_AT | OBD_CONNECT_OSS_CAPA |
472 OBD_CONNECT_VBR | OBD_CONNECT_FULL20 |
473 OBD_CONNECT_64BITHASH | OBD_CONNECT_MAXBYTES |
474 OBD_CONNECT_EINPROGRESS |
475 OBD_CONNECT_JOBSTATS | OBD_CONNECT_LVB_TYPE |
476 OBD_CONNECT_LAYOUTLOCK |
477 OBD_CONNECT_PINGLESS | OBD_CONNECT_LFSCK |
478 OBD_CONNECT_BULK_MBITS | OBD_CONNECT_SHORTIO |
479 OBD_CONNECT_FLAGS2 | OBD_CONNECT_GRANT_SHRINK;
480 data->ocd_connect_flags2 = OBD_CONNECT2_LOCKAHEAD |
481 OBD_CONNECT2_INC_XID | OBD_CONNECT2_LSEEK;
483 if (!OBD_FAIL_CHECK(OBD_FAIL_OSC_CONNECT_GRANT_PARAM))
484 data->ocd_connect_flags |= OBD_CONNECT_GRANT_PARAM;
486 /* OBD_CONNECT_CKSUM should always be set, even if checksums are
487 * disabled by default, because it can still be enabled on the
488 * fly via /sys. As a consequence, we still need to come to an
489 * agreement on the supported algorithms at connect time
491 data->ocd_connect_flags |= OBD_CONNECT_CKSUM;
493 if (OBD_FAIL_CHECK(OBD_FAIL_OSC_CKSUM_ADLER_ONLY))
494 data->ocd_cksum_types = OBD_CKSUM_ADLER;
496 data->ocd_cksum_types = obd_cksum_types_supported_client();
498 #ifdef HAVE_LRU_RESIZE_SUPPORT
499 data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
501 /* always ping even if server suppress_pings */
502 if (sbi->ll_flags & LL_SBI_ALWAYS_PING)
503 data->ocd_connect_flags &= ~OBD_CONNECT_PINGLESS;
505 if (ll_sbi_has_encrypt(sbi))
506 obd_connect_set_enc(data);
508 CDEBUG(D_RPCTRACE, "ocd_connect_flags: %#llx ocd_version: %d "
509 "ocd_grant: %d\n", data->ocd_connect_flags,
510 data->ocd_version, data->ocd_grant);
512 sbi->ll_dt_obd->obd_upcall.onu_owner = &sbi->ll_lco;
513 sbi->ll_dt_obd->obd_upcall.onu_upcall = cl_ocd_update;
515 data->ocd_brw_size = DT_MAX_BRW_SIZE;
517 err = obd_connect(NULL, &sbi->ll_dt_exp, sbi->ll_dt_obd,
518 &sbi->ll_sb_uuid, data, sbi->ll_cache);
520 LCONSOLE_ERROR_MSG(0x150, "An OST (dt %s) is performing "
521 "recovery, of which this client is not a "
522 "part. Please wait for recovery to "
523 "complete, abort, or time out.\n", dt);
526 CERROR("%s: Cannot connect to %s: rc = %d\n",
527 sbi->ll_dt_exp->exp_obd->obd_name, dt, err);
531 if (ll_sbi_has_encrypt(sbi) &&
532 !obd_connect_has_enc(&sbi->ll_dt_obd->u.lov.lov_ocd)) {
533 if (ll_sbi_has_test_dummy_encryption(sbi))
534 LCONSOLE_WARN("%s: server %s does not support encryption feature, encryption deactivated.\n",
536 ll_sbi_set_encrypt(sbi, false);
537 } else if (ll_sbi_has_test_dummy_encryption(sbi)) {
538 LCONSOLE_WARN("Test dummy encryption mode enabled\n");
541 sbi->ll_dt_exp->exp_connect_data = *data;
543 /* Don't change value if it was specified in the config log */
544 if (sbi->ll_ra_info.ra_max_read_ahead_whole_pages == -1) {
545 sbi->ll_ra_info.ra_max_read_ahead_whole_pages =
546 max_t(unsigned long, SBI_DEFAULT_READ_AHEAD_WHOLE_MAX,
547 (data->ocd_brw_size >> PAGE_SHIFT));
548 if (sbi->ll_ra_info.ra_max_read_ahead_whole_pages >
549 sbi->ll_ra_info.ra_max_pages_per_file)
550 sbi->ll_ra_info.ra_max_read_ahead_whole_pages =
551 sbi->ll_ra_info.ra_max_pages_per_file;
554 err = obd_fid_init(sbi->ll_dt_exp->exp_obd, sbi->ll_dt_exp,
555 LUSTRE_SEQ_METADATA);
557 CERROR("%s: Can't init data layer FID infrastructure, "
558 "rc = %d\n", sbi->ll_dt_exp->exp_obd->obd_name, err);
562 mutex_lock(&sbi->ll_lco.lco_lock);
563 sbi->ll_lco.lco_flags = data->ocd_connect_flags;
564 sbi->ll_lco.lco_md_exp = sbi->ll_md_exp;
565 sbi->ll_lco.lco_dt_exp = sbi->ll_dt_exp;
566 mutex_unlock(&sbi->ll_lco.lco_lock);
568 fid_zero(&sbi->ll_root_fid);
569 err = md_get_root(sbi->ll_md_exp, get_mount_fileset(sb),
572 CERROR("cannot mds_connect: rc = %d\n", err);
573 GOTO(out_lock_cn_cb, err);
575 if (!fid_is_sane(&sbi->ll_root_fid)) {
576 CERROR("%s: Invalid root fid "DFID" during mount\n",
577 sbi->ll_md_exp->exp_obd->obd_name,
578 PFID(&sbi->ll_root_fid));
579 GOTO(out_lock_cn_cb, err = -EINVAL);
581 CDEBUG(D_SUPER, "rootfid "DFID"\n", PFID(&sbi->ll_root_fid));
583 sb->s_op = &lustre_super_operations;
584 sb->s_xattr = ll_xattr_handlers;
585 #if THREAD_SIZE >= 8192 /*b=17630*/
586 sb->s_export_op = &lustre_export_operations;
588 #ifdef HAVE_LUSTRE_CRYPTO
589 llcrypt_set_ops(sb, &lustre_cryptops);
593 * XXX: move this to after cbd setup? */
594 valid = OBD_MD_FLGETATTR | OBD_MD_FLBLOCKS | OBD_MD_FLMODEASIZE;
595 if (sbi->ll_flags & LL_SBI_ACL)
596 valid |= OBD_MD_FLACL;
598 OBD_ALLOC_PTR(op_data);
600 GOTO(out_lock_cn_cb, err = -ENOMEM);
602 op_data->op_fid1 = sbi->ll_root_fid;
603 op_data->op_mode = 0;
604 op_data->op_valid = valid;
606 err = md_getattr(sbi->ll_md_exp, op_data, &request);
608 OBD_FREE_PTR(op_data);
610 CERROR("%s: md_getattr failed for root: rc = %d\n",
611 sbi->ll_md_exp->exp_obd->obd_name, err);
612 GOTO(out_lock_cn_cb, err);
615 err = md_get_lustre_md(sbi->ll_md_exp, request, sbi->ll_dt_exp,
616 sbi->ll_md_exp, &lmd);
618 CERROR("failed to understand root inode md: rc = %d\n", err);
619 ptlrpc_req_finished(request);
620 GOTO(out_lock_cn_cb, err);
623 LASSERT(fid_is_sane(&sbi->ll_root_fid));
624 root = ll_iget(sb, cl_fid_build_ino(&sbi->ll_root_fid,
625 sbi->ll_flags & LL_SBI_32BIT_API),
627 md_free_lustre_md(sbi->ll_md_exp, &lmd);
628 ptlrpc_req_finished(request);
632 err = IS_ERR(root) ? PTR_ERR(root) : -EBADF;
634 CERROR("%s: bad ll_iget() for root: rc = %d\n",
635 sbi->ll_fsname, err);
639 checksum = sbi->ll_flags & LL_SBI_CHECKSUM;
640 if (sbi->ll_checksum_set) {
641 err = obd_set_info_async(NULL, sbi->ll_dt_exp,
642 sizeof(KEY_CHECKSUM), KEY_CHECKSUM,
643 sizeof(checksum), &checksum, NULL);
645 CERROR("%s: Set checksum failed: rc = %d\n",
646 sbi->ll_dt_exp->exp_obd->obd_name, err);
652 sb->s_root = d_make_root(root);
653 if (sb->s_root == NULL) {
655 CERROR("%s: can't make root dentry: rc = %d\n",
656 sbi->ll_fsname, err);
660 sbi->ll_sdev_orig = sb->s_dev;
662 /* We set sb->s_dev equal on all lustre clients in order to support
663 * NFS export clustering. NFSD requires that the FSID be the same
665 /* s_dev is also used in lt_compare() to compare two fs, but that is
666 * only a node-local comparison. */
667 uuid = obd_get_uuid(sbi->ll_md_exp);
669 sb->s_dev = get_uuid2int(uuid->uuid, strlen(uuid->uuid));
676 if (sbi->ll_dt_obd) {
677 err = sysfs_create_link(&sbi->ll_kset.kobj,
678 &sbi->ll_dt_obd->obd_kset.kobj,
679 sbi->ll_dt_obd->obd_type->typ_name);
681 CERROR("%s: could not register %s in llite: rc = %d\n",
682 dt, sbi->ll_fsname, err);
687 if (sbi->ll_md_obd) {
688 err = sysfs_create_link(&sbi->ll_kset.kobj,
689 &sbi->ll_md_obd->obd_kset.kobj,
690 sbi->ll_md_obd->obd_type->typ_name);
692 CERROR("%s: could not register %s in llite: rc = %d\n",
693 md, sbi->ll_fsname, err);
703 obd_fid_fini(sbi->ll_dt_exp->exp_obd);
705 obd_disconnect(sbi->ll_dt_exp);
706 sbi->ll_dt_exp = NULL;
707 sbi->ll_dt_obd = NULL;
709 obd_fid_fini(sbi->ll_md_exp->exp_obd);
711 obd_disconnect(sbi->ll_md_exp);
712 sbi->ll_md_exp = NULL;
713 sbi->ll_md_obd = NULL;
722 int ll_get_max_mdsize(struct ll_sb_info *sbi, int *lmmsize)
726 size = sizeof(*lmmsize);
727 rc = obd_get_info(NULL, sbi->ll_dt_exp, sizeof(KEY_MAX_EASIZE),
728 KEY_MAX_EASIZE, &size, lmmsize);
730 CERROR("%s: cannot get max LOV EA size: rc = %d\n",
731 sbi->ll_dt_exp->exp_obd->obd_name, rc);
735 CDEBUG(D_INFO, "max LOV ea size: %d\n", *lmmsize);
738 rc = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_MAX_EASIZE),
739 KEY_MAX_EASIZE, &size, lmmsize);
741 CERROR("Get max mdsize error rc %d\n", rc);
743 CDEBUG(D_INFO, "max LMV ea size: %d\n", *lmmsize);
749 * Get the value of the default_easize parameter.
751 * \see client_obd::cl_default_mds_easize
753 * \param[in] sbi superblock info for this filesystem
754 * \param[out] lmmsize pointer to storage location for value
756 * \retval 0 on success
757 * \retval negative negated errno on failure
759 int ll_get_default_mdsize(struct ll_sb_info *sbi, int *lmmsize)
764 rc = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_DEFAULT_EASIZE),
765 KEY_DEFAULT_EASIZE, &size, lmmsize);
767 CERROR("Get default mdsize error rc %d\n", rc);
773 * Set the default_easize parameter to the given value.
775 * \see client_obd::cl_default_mds_easize
777 * \param[in] sbi superblock info for this filesystem
778 * \param[in] lmmsize the size to set
780 * \retval 0 on success
781 * \retval negative negated errno on failure
783 int ll_set_default_mdsize(struct ll_sb_info *sbi, int lmmsize)
787 if (lmmsize < sizeof(struct lov_mds_md) ||
788 lmmsize > OBD_MAX_DEFAULT_EA_SIZE)
791 rc = obd_set_info_async(NULL, sbi->ll_md_exp,
792 sizeof(KEY_DEFAULT_EASIZE), KEY_DEFAULT_EASIZE,
793 sizeof(int), &lmmsize, NULL);
798 static void client_common_put_super(struct super_block *sb)
800 struct ll_sb_info *sbi = ll_s2sbi(sb);
805 obd_fid_fini(sbi->ll_dt_exp->exp_obd);
806 obd_disconnect(sbi->ll_dt_exp);
807 sbi->ll_dt_exp = NULL;
809 ll_debugfs_unregister_super(sb);
811 obd_fid_fini(sbi->ll_md_exp->exp_obd);
812 obd_disconnect(sbi->ll_md_exp);
813 sbi->ll_md_exp = NULL;
818 void ll_kill_super(struct super_block *sb)
820 struct ll_sb_info *sbi;
824 if (!(sb->s_flags & SB_ACTIVE))
828 /* we need restore s_dev from changed for clustred NFS before put_super
829 * because new kernels have cached s_dev and change sb->s_dev in
830 * put_super not affected real removing devices */
832 sb->s_dev = sbi->ll_sdev_orig;
834 /* wait running statahead threads to quit */
835 while (atomic_read(&sbi->ll_sa_running) > 0)
836 schedule_timeout_uninterruptible(
837 cfs_time_seconds(1) >> 3);
843 static inline int ll_set_opt(const char *opt, char *data, int fl)
845 if (strncmp(opt, data, strlen(opt)) != 0)
851 /* non-client-specific mount options are parsed in lmd_parse */
852 static int ll_options(char *options, struct ll_sb_info *sbi)
855 char *s1 = options, *s2;
856 int *flags = &sbi->ll_flags;
862 CDEBUG(D_CONFIG, "Parsing opts %s\n", options);
865 CDEBUG(D_SUPER, "next opt=%s\n", s1);
866 tmp = ll_set_opt("nolock", s1, LL_SBI_NOLCK);
871 tmp = ll_set_opt("flock", s1, LL_SBI_FLOCK);
873 *flags = (*flags & ~LL_SBI_LOCALFLOCK) | tmp;
876 tmp = ll_set_opt("localflock", s1, LL_SBI_LOCALFLOCK);
878 *flags = (*flags & ~LL_SBI_FLOCK) | tmp;
881 tmp = ll_set_opt("noflock", s1, LL_SBI_FLOCK|LL_SBI_LOCALFLOCK);
886 tmp = ll_set_opt("user_xattr", s1, LL_SBI_USER_XATTR);
891 tmp = ll_set_opt("nouser_xattr", s1, LL_SBI_USER_XATTR);
896 tmp = ll_set_opt("context", s1, 1);
899 tmp = ll_set_opt("fscontext", s1, 1);
902 tmp = ll_set_opt("defcontext", s1, 1);
905 tmp = ll_set_opt("rootcontext", s1, 1);
908 tmp = ll_set_opt("user_fid2path", s1, LL_SBI_USER_FID2PATH);
913 tmp = ll_set_opt("nouser_fid2path", s1, LL_SBI_USER_FID2PATH);
919 tmp = ll_set_opt("checksum", s1, LL_SBI_CHECKSUM);
922 sbi->ll_checksum_set = 1;
925 tmp = ll_set_opt("nochecksum", s1, LL_SBI_CHECKSUM);
928 sbi->ll_checksum_set = 1;
931 tmp = ll_set_opt("lruresize", s1, LL_SBI_LRU_RESIZE);
936 tmp = ll_set_opt("nolruresize", s1, LL_SBI_LRU_RESIZE);
941 tmp = ll_set_opt("lazystatfs", s1, LL_SBI_LAZYSTATFS);
946 tmp = ll_set_opt("nolazystatfs", s1, LL_SBI_LAZYSTATFS);
951 tmp = ll_set_opt("32bitapi", s1, LL_SBI_32BIT_API);
956 tmp = ll_set_opt("verbose", s1, LL_SBI_VERBOSE);
961 tmp = ll_set_opt("noverbose", s1, LL_SBI_VERBOSE);
966 tmp = ll_set_opt("always_ping", s1, LL_SBI_ALWAYS_PING);
971 tmp = ll_set_opt("test_dummy_encryption", s1,
972 LL_SBI_TEST_DUMMY_ENCRYPTION);
974 #ifdef HAVE_LUSTRE_CRYPTO
977 LCONSOLE_WARN("Test dummy encryption mount option ignored: encryption not supported\n");
981 tmp = ll_set_opt("noencrypt", s1, LL_SBI_ENCRYPT);
983 #ifdef HAVE_LUSTRE_CRYPTO
986 LCONSOLE_WARN("noencrypt mount option ignored: encryption not supported\n");
990 LCONSOLE_ERROR_MSG(0x152, "Unknown option '%s', won't mount.\n",
996 s2 = strchr(s1, ',');
1004 void ll_lli_init(struct ll_inode_info *lli)
1006 lli->lli_inode_magic = LLI_INODE_MAGIC;
1008 spin_lock_init(&lli->lli_lock);
1009 lli->lli_posix_acl = NULL;
1010 /* Do not set lli_fid, it has been initialized already. */
1011 fid_zero(&lli->lli_pfid);
1012 lli->lli_mds_read_och = NULL;
1013 lli->lli_mds_write_och = NULL;
1014 lli->lli_mds_exec_och = NULL;
1015 lli->lli_open_fd_read_count = 0;
1016 lli->lli_open_fd_write_count = 0;
1017 lli->lli_open_fd_exec_count = 0;
1018 mutex_init(&lli->lli_och_mutex);
1019 spin_lock_init(&lli->lli_agl_lock);
1020 spin_lock_init(&lli->lli_layout_lock);
1021 ll_layout_version_set(lli, CL_LAYOUT_GEN_NONE);
1022 lli->lli_clob = NULL;
1024 init_rwsem(&lli->lli_xattrs_list_rwsem);
1025 mutex_init(&lli->lli_xattrs_enq_lock);
1027 LASSERT(lli->lli_vfs_inode.i_mode != 0);
1028 if (S_ISDIR(lli->lli_vfs_inode.i_mode)) {
1029 lli->lli_opendir_key = NULL;
1030 lli->lli_sai = NULL;
1031 spin_lock_init(&lli->lli_sa_lock);
1032 lli->lli_opendir_pid = 0;
1033 lli->lli_sa_enabled = 0;
1034 init_rwsem(&lli->lli_lsm_sem);
1036 mutex_init(&lli->lli_size_mutex);
1037 mutex_init(&lli->lli_setattr_mutex);
1038 lli->lli_symlink_name = NULL;
1039 ll_trunc_sem_init(&lli->lli_trunc_sem);
1040 range_lock_tree_init(&lli->lli_write_tree);
1041 init_rwsem(&lli->lli_glimpse_sem);
1042 lli->lli_glimpse_time = ktime_set(0, 0);
1043 INIT_LIST_HEAD(&lli->lli_agl_list);
1044 lli->lli_agl_index = 0;
1045 lli->lli_async_rc = 0;
1046 spin_lock_init(&lli->lli_heat_lock);
1047 obd_heat_clear(lli->lli_heat_instances, OBD_HEAT_COUNT);
1048 lli->lli_heat_flags = 0;
1049 mutex_init(&lli->lli_pcc_lock);
1050 lli->lli_pcc_state = PCC_STATE_FL_NONE;
1051 lli->lli_pcc_inode = NULL;
1052 lli->lli_pcc_dsflags = PCC_DATASET_INVALID;
1053 lli->lli_pcc_generation = 0;
1054 mutex_init(&lli->lli_group_mutex);
1055 lli->lli_group_users = 0;
1056 lli->lli_group_gid = 0;
1058 mutex_init(&lli->lli_layout_mutex);
1059 memset(lli->lli_jobid, 0, sizeof(lli->lli_jobid));
1062 #define MAX_STRING_SIZE 128
1064 #ifndef HAVE_SUPER_SETUP_BDI_NAME
1066 #define LSI_BDI_INITIALIZED 0x00400000
1068 #ifndef HAVE_BDI_CAP_MAP_COPY
1069 # define BDI_CAP_MAP_COPY 0
1072 static int super_setup_bdi_name(struct super_block *sb, char *fmt, ...)
1074 struct lustre_sb_info *lsi = s2lsi(sb);
1075 char buf[MAX_STRING_SIZE];
1079 err = bdi_init(&lsi->lsi_bdi);
1083 lsi->lsi_flags |= LSI_BDI_INITIALIZED;
1084 lsi->lsi_bdi.capabilities = BDI_CAP_MAP_COPY;
1085 lsi->lsi_bdi.name = "lustre";
1086 va_start(args, fmt);
1087 vsnprintf(buf, MAX_STRING_SIZE, fmt, args);
1089 err = bdi_register(&lsi->lsi_bdi, NULL, "%s", buf);
1092 sb->s_bdi = &lsi->lsi_bdi;
1096 #endif /* !HAVE_SUPER_SETUP_BDI_NAME */
1098 int ll_fill_super(struct super_block *sb)
1100 struct lustre_profile *lprof = NULL;
1101 struct lustre_sb_info *lsi = s2lsi(sb);
1102 struct ll_sb_info *sbi = NULL;
1103 char *dt = NULL, *md = NULL;
1104 char *profilenm = get_profile_name(sb);
1105 struct config_llog_instance *cfg;
1106 /* %p for void* in printf needs 16+2 characters: 0xffffffffffffffff */
1107 const int instlen = LUSTRE_MAXINSTANCE + 2;
1108 unsigned long cfg_instance = ll_get_cfg_instance(sb);
1109 char name[MAX_STRING_SIZE];
1118 /* for ASLR, to map between cfg_instance and hashed ptr */
1119 CDEBUG(D_VFSTRACE, "VFS Op: cfg_instance %s-%016lx (sb %p)\n",
1120 profilenm, cfg_instance, sb);
1124 GOTO(out_free_cfg, err = -ENOMEM);
1126 /* client additional sb info */
1127 lsi->lsi_llsbi = sbi = ll_init_sbi();
1129 GOTO(out_free_cfg, err = PTR_ERR(sbi));
1131 err = ll_options(lsi->lsi_lmd->lmd_opts, sbi);
1133 GOTO(out_free_cfg, err);
1135 /* kernel >= 2.6.38 store dentry operations in sb->s_d_op. */
1136 sb->s_d_op = &ll_d_ops;
1139 generate_random_uuid(uuid.b);
1140 snprintf(sbi->ll_sb_uuid.uuid, sizeof(sbi->ll_sb_uuid), "%pU", uuid.b);
1142 CDEBUG(D_CONFIG, "llite sb uuid: %s\n", sbi->ll_sb_uuid.uuid);
1145 len = strlen(profilenm);
1146 ptr = strrchr(profilenm, '-');
1147 if (ptr && (strcmp(ptr, "-client") == 0))
1150 if (len > LUSTRE_MAXFSNAME) {
1151 if (unlikely(len >= MAX_STRING_SIZE))
1152 len = MAX_STRING_SIZE - 1;
1153 strncpy(name, profilenm, len);
1155 err = -ENAMETOOLONG;
1156 CERROR("%s: fsname longer than %u characters: rc = %d\n",
1157 name, LUSTRE_MAXFSNAME, err);
1158 GOTO(out_free_cfg, err);
1160 strncpy(sbi->ll_fsname, profilenm, len);
1161 sbi->ll_fsname[len] = '\0';
1164 snprintf(name, sizeof(name), "%.*s-%016lx", len,
1165 profilenm, cfg_instance);
1167 err = super_setup_bdi_name(sb, "%s", name);
1169 GOTO(out_free_cfg, err);
1171 /* Call ll_debugfs_register_super() before lustre_process_log()
1172 * so that "llite.*.*" params can be processed correctly.
1174 err = ll_debugfs_register_super(sb, name);
1176 CERROR("%s: could not register mountpoint in llite: rc = %d\n",
1177 sbi->ll_fsname, err);
1181 /* The cfg_instance is a value unique to this super, in case some
1182 * joker tries to mount the same fs at two mount points.
1184 cfg->cfg_instance = cfg_instance;
1185 cfg->cfg_uuid = lsi->lsi_llsbi->ll_sb_uuid;
1186 cfg->cfg_callback = class_config_llog_handler;
1187 cfg->cfg_sub_clds = CONFIG_SUB_CLIENT;
1188 /* set up client obds */
1189 err = lustre_process_log(sb, profilenm, cfg);
1191 GOTO(out_debugfs, err);
1193 /* Profile set with LCFG_MOUNTOPT so we can find our mdc and osc obds */
1194 lprof = class_get_profile(profilenm);
1195 if (lprof == NULL) {
1196 LCONSOLE_ERROR_MSG(0x156, "The client profile '%s' could not be"
1197 " read from the MGS. Does that filesystem "
1198 "exist?\n", profilenm);
1199 GOTO(out_debugfs, err = -EINVAL);
1201 CDEBUG(D_CONFIG, "Found profile %s: mdc=%s osc=%s\n", profilenm,
1202 lprof->lp_md, lprof->lp_dt);
1204 dt_len = strlen(lprof->lp_dt) + instlen + 2;
1205 OBD_ALLOC(dt, dt_len);
1207 GOTO(out_profile, err = -ENOMEM);
1208 snprintf(dt, dt_len - 1, "%s-%016lx", lprof->lp_dt, cfg_instance);
1210 md_len = strlen(lprof->lp_md) + instlen + 2;
1211 OBD_ALLOC(md, md_len);
1213 GOTO(out_free_dt, err = -ENOMEM);
1214 snprintf(md, md_len - 1, "%s-%016lx", lprof->lp_md, cfg_instance);
1216 /* connections, registrations, sb setup */
1217 err = client_common_fill_super(sb, md, dt);
1219 GOTO(out_free_md, err);
1221 sbi->ll_client_common_fill_super_succeeded = 1;
1225 OBD_FREE(md, md_len);
1228 OBD_FREE(dt, dt_len);
1231 class_put_profile(lprof);
1234 ll_debugfs_unregister_super(sb);
1241 else if (sbi->ll_flags & LL_SBI_VERBOSE)
1242 LCONSOLE_WARN("Mounted %s\n", profilenm);
1244 } /* ll_fill_super */
1246 void ll_put_super(struct super_block *sb)
1248 struct config_llog_instance cfg, params_cfg;
1249 struct obd_device *obd;
1250 struct lustre_sb_info *lsi = s2lsi(sb);
1251 struct ll_sb_info *sbi = ll_s2sbi(sb);
1252 char *profilenm = get_profile_name(sb);
1253 unsigned long cfg_instance = ll_get_cfg_instance(sb);
1255 int next, force = 1, rc = 0;
1259 GOTO(out_no_sbi, 0);
1261 /* Should replace instance_id with something better for ASLR */
1262 CDEBUG(D_VFSTRACE, "VFS Op: cfg_instance %s-%016lx (sb %p)\n",
1263 profilenm, cfg_instance, sb);
1265 cfg.cfg_instance = cfg_instance;
1266 lustre_end_log(sb, profilenm, &cfg);
1268 params_cfg.cfg_instance = cfg_instance;
1269 lustre_end_log(sb, PARAMS_FILENAME, ¶ms_cfg);
1271 if (sbi->ll_md_exp) {
1272 obd = class_exp2obd(sbi->ll_md_exp);
1274 force = obd->obd_force;
1277 /* Wait for unstable pages to be committed to stable storage */
1279 rc = l_wait_event_abortable(
1280 sbi->ll_cache->ccc_unstable_waitq,
1281 atomic_long_read(&sbi->ll_cache->ccc_unstable_nr) == 0);
1284 ccc_count = atomic_long_read(&sbi->ll_cache->ccc_unstable_nr);
1285 if (force == 0 && rc != -ERESTARTSYS)
1286 LASSERTF(ccc_count == 0, "count: %li\n", ccc_count);
1288 /* We need to set force before the lov_disconnect in
1289 * lustre_common_put_super, since l_d cleans up osc's as well.
1293 while ((obd = class_devices_in_group(&sbi->ll_sb_uuid,
1295 obd->obd_force = force;
1299 if (sbi->ll_client_common_fill_super_succeeded) {
1300 /* Only if client_common_fill_super succeeded */
1301 client_common_put_super(sb);
1305 while ((obd = class_devices_in_group(&sbi->ll_sb_uuid, &next)))
1306 class_manual_cleanup(obd);
1308 if (sbi->ll_flags & LL_SBI_VERBOSE)
1309 LCONSOLE_WARN("Unmounted %s\n", profilenm ? profilenm : "");
1312 class_del_profile(profilenm);
1314 #ifndef HAVE_SUPER_SETUP_BDI_NAME
1315 if (lsi->lsi_flags & LSI_BDI_INITIALIZED) {
1316 bdi_destroy(&lsi->lsi_bdi);
1317 lsi->lsi_flags &= ~LSI_BDI_INITIALIZED;
1322 lsi->lsi_llsbi = NULL;
1324 lustre_common_put_super(sb);
1326 cl_env_cache_purge(~0);
1328 module_put(THIS_MODULE);
1331 } /* client_put_super */
1333 struct inode *ll_inode_from_resource_lock(struct ldlm_lock *lock)
1335 struct inode *inode = NULL;
1337 /* NOTE: we depend on atomic igrab() -bzzz */
1338 lock_res_and_lock(lock);
1339 if (lock->l_resource->lr_lvb_inode) {
1340 struct ll_inode_info * lli;
1341 lli = ll_i2info(lock->l_resource->lr_lvb_inode);
1342 if (lli->lli_inode_magic == LLI_INODE_MAGIC) {
1343 inode = igrab(lock->l_resource->lr_lvb_inode);
1345 inode = lock->l_resource->lr_lvb_inode;
1346 LDLM_DEBUG_LIMIT(inode->i_state & I_FREEING ? D_INFO :
1347 D_WARNING, lock, "lr_lvb_inode %p is "
1348 "bogus: magic %08x",
1349 lock->l_resource->lr_lvb_inode,
1350 lli->lli_inode_magic);
1354 unlock_res_and_lock(lock);
1358 void ll_dir_clear_lsm_md(struct inode *inode)
1360 struct ll_inode_info *lli = ll_i2info(inode);
1362 LASSERT(S_ISDIR(inode->i_mode));
1364 if (lli->lli_lsm_md) {
1365 lmv_free_memmd(lli->lli_lsm_md);
1366 lli->lli_lsm_md = NULL;
1369 if (lli->lli_default_lsm_md) {
1370 lmv_free_memmd(lli->lli_default_lsm_md);
1371 lli->lli_default_lsm_md = NULL;
1375 static struct inode *ll_iget_anon_dir(struct super_block *sb,
1376 const struct lu_fid *fid,
1377 struct lustre_md *md)
1379 struct ll_sb_info *sbi = ll_s2sbi(sb);
1380 struct mdt_body *body = md->body;
1381 struct inode *inode;
1385 ino = cl_fid_build_ino(fid, sbi->ll_flags & LL_SBI_32BIT_API);
1386 inode = iget_locked(sb, ino);
1387 if (inode == NULL) {
1388 CERROR("%s: failed get simple inode "DFID": rc = -ENOENT\n",
1389 sbi->ll_fsname, PFID(fid));
1390 RETURN(ERR_PTR(-ENOENT));
1393 if (inode->i_state & I_NEW) {
1394 struct ll_inode_info *lli = ll_i2info(inode);
1395 struct lmv_stripe_md *lsm = md->lmv;
1397 inode->i_mode = (inode->i_mode & ~S_IFMT) |
1398 (body->mbo_mode & S_IFMT);
1399 LASSERTF(S_ISDIR(inode->i_mode), "Not slave inode "DFID"\n",
1402 inode->i_mtime.tv_sec = 0;
1403 inode->i_atime.tv_sec = 0;
1404 inode->i_ctime.tv_sec = 0;
1407 #ifdef HAVE_BACKING_DEV_INFO
1408 /* initializing backing dev info. */
1409 inode->i_mapping->backing_dev_info =
1410 &s2lsi(inode->i_sb)->lsi_bdi;
1412 inode->i_op = &ll_dir_inode_operations;
1413 inode->i_fop = &ll_dir_operations;
1414 lli->lli_fid = *fid;
1417 LASSERT(lsm != NULL);
1418 /* master object FID */
1419 lli->lli_pfid = body->mbo_fid1;
1420 CDEBUG(D_INODE, "lli %p slave "DFID" master "DFID"\n",
1421 lli, PFID(fid), PFID(&lli->lli_pfid));
1422 unlock_new_inode(inode);
1428 static int ll_init_lsm_md(struct inode *inode, struct lustre_md *md)
1431 struct lmv_stripe_md *lsm = md->lmv;
1432 struct ll_inode_info *lli = ll_i2info(inode);
1435 LASSERT(lsm != NULL);
1437 CDEBUG(D_INODE, "%s: "DFID" set dir layout:\n",
1438 ll_i2sbi(inode)->ll_fsname, PFID(&lli->lli_fid));
1439 lsm_md_dump(D_INODE, lsm);
1441 if (!lmv_dir_striped(lsm))
1444 /* XXX sigh, this lsm_root initialization should be in
1445 * LMV layer, but it needs ll_iget right now, so we
1446 * put this here right now. */
1447 for (i = 0; i < lsm->lsm_md_stripe_count; i++) {
1448 fid = &lsm->lsm_md_oinfo[i].lmo_fid;
1449 LASSERT(lsm->lsm_md_oinfo[i].lmo_root == NULL);
1451 if (!fid_is_sane(fid))
1454 /* Unfortunately ll_iget will call ll_update_inode,
1455 * where the initialization of slave inode is slightly
1456 * different, so it reset lsm_md to NULL to avoid
1457 * initializing lsm for slave inode. */
1458 lsm->lsm_md_oinfo[i].lmo_root =
1459 ll_iget_anon_dir(inode->i_sb, fid, md);
1460 if (IS_ERR(lsm->lsm_md_oinfo[i].lmo_root)) {
1461 int rc = PTR_ERR(lsm->lsm_md_oinfo[i].lmo_root);
1463 lsm->lsm_md_oinfo[i].lmo_root = NULL;
1465 iput(lsm->lsm_md_oinfo[i].lmo_root);
1466 lsm->lsm_md_oinfo[i].lmo_root = NULL;
1472 lli->lli_lsm_md = lsm;
1477 static void ll_update_default_lsm_md(struct inode *inode, struct lustre_md *md)
1479 struct ll_inode_info *lli = ll_i2info(inode);
1481 if (!md->default_lmv) {
1482 /* clear default lsm */
1483 if (lli->lli_default_lsm_md) {
1484 down_write(&lli->lli_lsm_sem);
1485 if (lli->lli_default_lsm_md) {
1486 lmv_free_memmd(lli->lli_default_lsm_md);
1487 lli->lli_default_lsm_md = NULL;
1489 up_write(&lli->lli_lsm_sem);
1491 } else if (lli->lli_default_lsm_md) {
1492 /* update default lsm if it changes */
1493 down_read(&lli->lli_lsm_sem);
1494 if (lli->lli_default_lsm_md &&
1495 !lsm_md_eq(lli->lli_default_lsm_md, md->default_lmv)) {
1496 up_read(&lli->lli_lsm_sem);
1497 down_write(&lli->lli_lsm_sem);
1498 if (lli->lli_default_lsm_md)
1499 lmv_free_memmd(lli->lli_default_lsm_md);
1500 lli->lli_default_lsm_md = md->default_lmv;
1501 lsm_md_dump(D_INODE, md->default_lmv);
1502 md->default_lmv = NULL;
1503 up_write(&lli->lli_lsm_sem);
1505 up_read(&lli->lli_lsm_sem);
1508 /* init default lsm */
1509 down_write(&lli->lli_lsm_sem);
1510 lli->lli_default_lsm_md = md->default_lmv;
1511 lsm_md_dump(D_INODE, md->default_lmv);
1512 md->default_lmv = NULL;
1513 up_write(&lli->lli_lsm_sem);
1517 static int ll_update_lsm_md(struct inode *inode, struct lustre_md *md)
1519 struct ll_inode_info *lli = ll_i2info(inode);
1520 struct lmv_stripe_md *lsm = md->lmv;
1521 struct cl_attr *attr;
1526 LASSERT(S_ISDIR(inode->i_mode));
1527 CDEBUG(D_INODE, "update lsm %p of "DFID"\n", lli->lli_lsm_md,
1528 PFID(ll_inode2fid(inode)));
1530 /* update default LMV */
1531 if (md->default_lmv)
1532 ll_update_default_lsm_md(inode, md);
1535 * no striped information from request, lustre_md from req does not
1536 * include stripeEA, see ll_md_setattr()
1542 * normally dir layout doesn't change, only take read lock to check
1543 * that to avoid blocking other MD operations.
1545 down_read(&lli->lli_lsm_sem);
1547 /* some current lookup initialized lsm, and unchanged */
1548 if (lli->lli_lsm_md && lsm_md_eq(lli->lli_lsm_md, lsm))
1549 GOTO(unlock, rc = 0);
1551 /* if dir layout doesn't match, check whether version is increased,
1552 * which means layout is changed, this happens in dir split/merge and
1555 * foreign LMV should not change.
1557 if (lli->lli_lsm_md && lmv_dir_striped(lli->lli_lsm_md) &&
1558 lsm->lsm_md_layout_version <=
1559 lli->lli_lsm_md->lsm_md_layout_version) {
1560 CERROR("%s: "DFID" dir layout mismatch:\n",
1561 ll_i2sbi(inode)->ll_fsname, PFID(&lli->lli_fid));
1562 lsm_md_dump(D_ERROR, lli->lli_lsm_md);
1563 lsm_md_dump(D_ERROR, lsm);
1564 GOTO(unlock, rc = -EINVAL);
1567 up_read(&lli->lli_lsm_sem);
1568 down_write(&lli->lli_lsm_sem);
1569 /* clear existing lsm */
1570 if (lli->lli_lsm_md) {
1571 lmv_free_memmd(lli->lli_lsm_md);
1572 lli->lli_lsm_md = NULL;
1575 rc = ll_init_lsm_md(inode, md);
1576 up_write(&lli->lli_lsm_sem);
1581 /* set md->lmv to NULL, so the following free lustre_md will not free
1586 /* md_merge_attr() may take long, since lsm is already set, switch to
1589 down_read(&lli->lli_lsm_sem);
1591 if (!lmv_dir_striped(lli->lli_lsm_md))
1592 GOTO(unlock, rc = 0);
1594 OBD_ALLOC_PTR(attr);
1596 GOTO(unlock, rc = -ENOMEM);
1598 /* validate the lsm */
1599 rc = md_merge_attr(ll_i2mdexp(inode), &lli->lli_fid, lli->lli_lsm_md,
1600 attr, ll_md_blocking_ast);
1602 if (md->body->mbo_valid & OBD_MD_FLNLINK)
1603 md->body->mbo_nlink = attr->cat_nlink;
1604 if (md->body->mbo_valid & OBD_MD_FLSIZE)
1605 md->body->mbo_size = attr->cat_size;
1606 if (md->body->mbo_valid & OBD_MD_FLATIME)
1607 md->body->mbo_atime = attr->cat_atime;
1608 if (md->body->mbo_valid & OBD_MD_FLCTIME)
1609 md->body->mbo_ctime = attr->cat_ctime;
1610 if (md->body->mbo_valid & OBD_MD_FLMTIME)
1611 md->body->mbo_mtime = attr->cat_mtime;
1617 up_read(&lli->lli_lsm_sem);
1622 void ll_clear_inode(struct inode *inode)
1624 struct ll_inode_info *lli = ll_i2info(inode);
1625 struct ll_sb_info *sbi = ll_i2sbi(inode);
1629 CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p)\n",
1630 PFID(ll_inode2fid(inode)), inode);
1632 if (S_ISDIR(inode->i_mode)) {
1633 /* these should have been cleared in ll_file_release */
1634 LASSERT(lli->lli_opendir_key == NULL);
1635 LASSERT(lli->lli_sai == NULL);
1636 LASSERT(lli->lli_opendir_pid == 0);
1638 pcc_inode_free(inode);
1641 md_null_inode(sbi->ll_md_exp, ll_inode2fid(inode));
1643 LASSERT(!lli->lli_open_fd_write_count);
1644 LASSERT(!lli->lli_open_fd_read_count);
1645 LASSERT(!lli->lli_open_fd_exec_count);
1647 if (lli->lli_mds_write_och)
1648 ll_md_real_close(inode, FMODE_WRITE);
1649 if (lli->lli_mds_exec_och)
1650 ll_md_real_close(inode, FMODE_EXEC);
1651 if (lli->lli_mds_read_och)
1652 ll_md_real_close(inode, FMODE_READ);
1654 if (S_ISLNK(inode->i_mode) && lli->lli_symlink_name) {
1655 OBD_FREE(lli->lli_symlink_name,
1656 strlen(lli->lli_symlink_name) + 1);
1657 lli->lli_symlink_name = NULL;
1660 ll_xattr_cache_destroy(inode);
1662 forget_all_cached_acls(inode);
1664 lli->lli_inode_magic = LLI_INODE_DEAD;
1666 if (S_ISDIR(inode->i_mode))
1667 ll_dir_clear_lsm_md(inode);
1668 else if (S_ISREG(inode->i_mode) && !is_bad_inode(inode))
1669 LASSERT(list_empty(&lli->lli_agl_list));
1672 * XXX This has to be done before lsm is freed below, because
1673 * cl_object still uses inode lsm.
1675 cl_inode_fini(inode);
1677 llcrypt_put_encryption_info(inode);
1682 static int ll_md_setattr(struct dentry *dentry, struct md_op_data *op_data)
1684 struct lustre_md md;
1685 struct inode *inode = dentry->d_inode;
1686 struct ll_sb_info *sbi = ll_i2sbi(inode);
1687 struct ptlrpc_request *request = NULL;
1691 op_data = ll_prep_md_op_data(op_data, inode, NULL, NULL, 0, 0,
1692 LUSTRE_OPC_ANY, NULL);
1693 if (IS_ERR(op_data))
1694 RETURN(PTR_ERR(op_data));
1696 rc = md_setattr(sbi->ll_md_exp, op_data, NULL, 0, &request);
1698 ptlrpc_req_finished(request);
1699 if (rc == -ENOENT) {
1701 /* Unlinked special device node? Or just a race?
1702 * Pretend we done everything. */
1703 if (!S_ISREG(inode->i_mode) &&
1704 !S_ISDIR(inode->i_mode)) {
1705 ia_valid = op_data->op_attr.ia_valid;
1706 op_data->op_attr.ia_valid &= ~TIMES_SET_FLAGS;
1707 rc = simple_setattr(dentry, &op_data->op_attr);
1708 op_data->op_attr.ia_valid = ia_valid;
1710 } else if (rc != -EPERM && rc != -EACCES && rc != -ETXTBSY) {
1711 CERROR("md_setattr fails: rc = %d\n", rc);
1716 rc = md_get_lustre_md(sbi->ll_md_exp, request, sbi->ll_dt_exp,
1717 sbi->ll_md_exp, &md);
1719 ptlrpc_req_finished(request);
1723 ia_valid = op_data->op_attr.ia_valid;
1724 /* inode size will be in ll_setattr_ost, can't do it now since dirty
1725 * cache is not cleared yet. */
1726 op_data->op_attr.ia_valid &= ~(TIMES_SET_FLAGS | ATTR_SIZE);
1727 if (S_ISREG(inode->i_mode))
1729 rc = simple_setattr(dentry, &op_data->op_attr);
1730 if (S_ISREG(inode->i_mode))
1731 inode_unlock(inode);
1732 op_data->op_attr.ia_valid = ia_valid;
1734 rc = ll_update_inode(inode, &md);
1735 ptlrpc_req_finished(request);
1741 * Zero portion of page that is part of @inode.
1742 * This implies, if necessary:
1743 * - taking cl_lock on range corresponding to concerned page
1744 * - grabbing vm page
1745 * - associating cl_page
1746 * - proceeding to clio read
1747 * - zeroing range in page
1748 * - proceeding to cl_page flush
1749 * - releasing cl_lock
1751 * \param[in] inode inode
1752 * \param[in] index page index
1753 * \param[in] offset offset in page to start zero from
1754 * \param[in] len len to zero
1756 * \retval 0 on success
1757 * \retval negative errno on failure
1759 int ll_io_zero_page(struct inode *inode, pgoff_t index, pgoff_t offset,
1762 struct ll_inode_info *lli = ll_i2info(inode);
1763 struct cl_object *clob = lli->lli_clob;
1765 struct lu_env *env = NULL;
1766 struct cl_io *io = NULL;
1767 struct cl_page *clpage = NULL;
1768 struct page *vmpage = NULL;
1769 unsigned from = index << PAGE_SHIFT;
1770 struct cl_lock *lock = NULL;
1771 struct cl_lock_descr *descr = NULL;
1772 struct cl_2queue *queue = NULL;
1773 struct cl_sync_io *anchor = NULL;
1774 bool holdinglock = false;
1775 bool lockedbymyself = true;
1780 env = cl_env_get(&refcheck);
1782 RETURN(PTR_ERR(env));
1784 io = vvp_env_thread_io(env);
1786 rc = cl_io_rw_init(env, io, CIT_WRITE, from, PAGE_SIZE);
1790 lock = vvp_env_lock(env);
1791 descr = &lock->cll_descr;
1792 descr->cld_obj = io->ci_obj;
1793 descr->cld_start = cl_index(io->ci_obj, from);
1794 descr->cld_end = cl_index(io->ci_obj, from + PAGE_SIZE - 1);
1795 descr->cld_mode = CLM_WRITE;
1796 descr->cld_enq_flags = CEF_MUST | CEF_NONBLOCK;
1798 /* request lock for page */
1799 rc = cl_lock_request(env, io, lock);
1800 /* -ECANCELED indicates a matching lock with a different extent
1801 * was already present, and -EEXIST indicates a matching lock
1802 * on exactly the same extent was already present.
1803 * In both cases it means we are covered.
1805 if (rc == -ECANCELED || rc == -EEXIST)
1813 vmpage = grab_cache_page_nowait(inode->i_mapping, index);
1815 GOTO(rellock, rc = -EOPNOTSUPP);
1817 if (!PageDirty(vmpage)) {
1818 /* associate cl_page */
1819 clpage = cl_page_find(env, clob, vmpage->index,
1820 vmpage, CPT_CACHEABLE);
1822 GOTO(pagefini, rc = PTR_ERR(clpage));
1824 cl_page_assume(env, io, clpage);
1827 if (!PageUptodate(vmpage) && !PageDirty(vmpage) &&
1828 !PageWriteback(vmpage)) {
1830 /* set PagePrivate2 to detect special case of empty page
1831 * in osc_brw_fini_request()
1833 SetPagePrivate2(vmpage);
1834 rc = ll_io_read_page(env, io, clpage, NULL);
1835 if (!PagePrivate2(vmpage))
1836 /* PagePrivate2 was cleared in osc_brw_fini_request()
1837 * meaning we read an empty page. In this case, in order
1838 * to avoid allocating unnecessary block in truncated
1839 * file, we must not zero and write as below. Subsequent
1840 * server-side truncate will handle things correctly.
1842 GOTO(clpfini, rc = 0);
1843 ClearPagePrivate2(vmpage);
1846 lockedbymyself = trylock_page(vmpage);
1847 cl_page_assume(env, io, clpage);
1850 /* zero range in page */
1851 zero_user(vmpage, offset, len);
1853 if (holdinglock && clpage) {
1854 /* explicitly write newly modified page */
1855 queue = &io->ci_queue;
1856 cl_2queue_init(queue);
1857 anchor = &vvp_env_info(env)->vti_anchor;
1858 cl_sync_io_init(anchor, 1);
1859 clpage->cp_sync_io = anchor;
1860 cl_2queue_add(queue, clpage);
1861 rc = cl_io_submit_rw(env, io, CRT_WRITE, queue);
1863 GOTO(queuefini1, rc);
1864 rc = cl_sync_io_wait(env, anchor, 0);
1866 GOTO(queuefini2, rc);
1867 cl_page_assume(env, io, clpage);
1870 cl_2queue_discard(env, io, queue);
1872 cl_2queue_disown(env, io, queue);
1873 cl_2queue_fini(env, queue);
1878 cl_page_put(env, clpage);
1880 if (lockedbymyself) {
1881 unlock_page(vmpage);
1886 cl_lock_release(env, lock);
1888 cl_io_fini(env, io);
1891 cl_env_put(env, &refcheck);
1896 /* If this inode has objects allocated to it (lsm != NULL), then the OST
1897 * object(s) determine the file size and mtime. Otherwise, the MDS will
1898 * keep these values until such a time that objects are allocated for it.
1899 * We do the MDS operations first, as it is checking permissions for us.
1900 * We don't to the MDS RPC if there is nothing that we want to store there,
1901 * otherwise there is no harm in updating mtime/atime on the MDS if we are
1902 * going to do an RPC anyways.
1904 * If we are doing a truncate, we will send the mtime and ctime updates
1905 * to the OST with the punch RPC, otherwise we do an explicit setattr RPC.
1906 * I don't believe it is possible to get e.g. ATTR_MTIME_SET and ATTR_SIZE
1909 * In case of HSMimport, we only set attr on MDS.
1911 int ll_setattr_raw(struct dentry *dentry, struct iattr *attr,
1912 enum op_xvalid xvalid, bool hsm_import)
1914 struct inode *inode = dentry->d_inode;
1915 struct ll_inode_info *lli = ll_i2info(inode);
1916 struct md_op_data *op_data = NULL;
1917 ktime_t kstart = ktime_get();
1922 CDEBUG(D_VFSTRACE, "%s: setattr inode "DFID"(%p) from %llu to %llu, "
1923 "valid %x, hsm_import %d\n",
1924 ll_i2sbi(inode)->ll_fsname, PFID(&lli->lli_fid),
1925 inode, i_size_read(inode), attr->ia_size, attr->ia_valid,
1928 if (attr->ia_valid & ATTR_SIZE) {
1929 /* Check new size against VFS/VM file size limit and rlimit */
1930 rc = inode_newsize_ok(inode, attr->ia_size);
1934 /* The maximum Lustre file size is variable, based on the
1935 * OST maximum object size and number of stripes. This
1936 * needs another check in addition to the VFS check above. */
1937 if (attr->ia_size > ll_file_maxbytes(inode)) {
1938 CDEBUG(D_INODE,"file "DFID" too large %llu > %llu\n",
1939 PFID(&lli->lli_fid), attr->ia_size,
1940 ll_file_maxbytes(inode));
1944 attr->ia_valid |= ATTR_MTIME | ATTR_CTIME;
1947 /* POSIX: check before ATTR_*TIME_SET set (from inode_change_ok) */
1948 if (attr->ia_valid & TIMES_SET_FLAGS) {
1949 if ((!uid_eq(current_fsuid(), inode->i_uid)) &&
1950 !cfs_capable(CFS_CAP_FOWNER))
1954 /* We mark all of the fields "set" so MDS/OST does not re-set them */
1955 if (!(xvalid & OP_XVALID_CTIME_SET) &&
1956 (attr->ia_valid & ATTR_CTIME)) {
1957 attr->ia_ctime = current_time(inode);
1958 xvalid |= OP_XVALID_CTIME_SET;
1960 if (!(attr->ia_valid & ATTR_ATIME_SET) &&
1961 (attr->ia_valid & ATTR_ATIME)) {
1962 attr->ia_atime = current_time(inode);
1963 attr->ia_valid |= ATTR_ATIME_SET;
1965 if (!(attr->ia_valid & ATTR_MTIME_SET) &&
1966 (attr->ia_valid & ATTR_MTIME)) {
1967 attr->ia_mtime = current_time(inode);
1968 attr->ia_valid |= ATTR_MTIME_SET;
1971 if (attr->ia_valid & (ATTR_MTIME | ATTR_CTIME))
1972 CDEBUG(D_INODE, "setting mtime %lld, ctime %lld, now = %lld\n",
1973 (s64)attr->ia_mtime.tv_sec, (s64)attr->ia_ctime.tv_sec,
1974 ktime_get_real_seconds());
1976 if (S_ISREG(inode->i_mode))
1977 inode_unlock(inode);
1979 /* We always do an MDS RPC, even if we're only changing the size;
1980 * only the MDS knows whether truncate() should fail with -ETXTBUSY */
1982 OBD_ALLOC_PTR(op_data);
1983 if (op_data == NULL)
1984 GOTO(out, rc = -ENOMEM);
1986 if (!hsm_import && attr->ia_valid & ATTR_SIZE) {
1987 /* If we are changing file size, file content is
1988 * modified, flag it.
1990 xvalid |= OP_XVALID_OWNEROVERRIDE;
1991 op_data->op_bias |= MDS_DATA_MODIFIED;
1992 ll_file_clear_flag(lli, LLIF_DATA_MODIFIED);
1995 if (attr->ia_valid & ATTR_FILE) {
1996 struct ll_file_data *fd = attr->ia_file->private_data;
1998 if (fd->fd_lease_och)
1999 op_data->op_bias |= MDS_TRUNC_KEEP_LEASE;
2002 op_data->op_attr = *attr;
2003 op_data->op_xvalid = xvalid;
2005 rc = ll_md_setattr(dentry, op_data);
2009 if (!S_ISREG(inode->i_mode) || hsm_import)
2012 if (attr->ia_valid & (ATTR_SIZE | ATTR_ATIME | ATTR_ATIME_SET |
2013 ATTR_MTIME | ATTR_MTIME_SET | ATTR_CTIME) ||
2014 xvalid & OP_XVALID_CTIME_SET) {
2015 bool cached = false;
2017 rc = pcc_inode_setattr(inode, attr, &cached);
2020 CERROR("%s: PCC inode "DFID" setattr failed: "
2022 ll_i2sbi(inode)->ll_fsname,
2023 PFID(&lli->lli_fid), rc);
2027 unsigned int flags = 0;
2029 /* For truncate and utimes sending attributes to OSTs,
2030 * setting mtime/atime to the past will be performed
2031 * under PW [0:EOF] extent lock (new_size:EOF for
2032 * truncate). It may seem excessive to send mtime/atime
2033 * updates to OSTs when not setting times to past, but
2034 * it is necessary due to possible time
2035 * de-synchronization between MDT inode and OST objects
2037 if (S_ISREG(inode->i_mode) && IS_ENCRYPTED(inode) &&
2038 attr->ia_valid & ATTR_SIZE) {
2039 xvalid |= OP_XVALID_FLAGS;
2040 flags = LUSTRE_ENCRYPT_FL;
2041 /* Call to ll_io_zero_page is not necessary if
2042 * truncating on PAGE_SIZE boundary, because
2043 * whole pages will be wiped.
2044 * In case of Direct IO, all we need is to set
2047 if (attr->ia_size & ~PAGE_MASK &&
2048 !(attr->ia_valid & ATTR_FILE &&
2049 attr->ia_file->f_flags & O_DIRECT)) {
2051 attr->ia_size & (PAGE_SIZE - 1);
2053 rc = ll_io_zero_page(inode,
2054 attr->ia_size >> PAGE_SHIFT,
2055 offset, PAGE_SIZE - offset);
2060 rc = cl_setattr_ost(lli->lli_clob, attr, xvalid, flags);
2064 /* If the file was restored, it needs to set dirty flag.
2066 * We've already sent MDS_DATA_MODIFIED flag in
2067 * ll_md_setattr() for truncate. However, the MDT refuses to
2068 * set the HS_DIRTY flag on released files, so we have to set
2069 * it again if the file has been restored. Please check how
2070 * LLIF_DATA_MODIFIED is set in vvp_io_setattr_fini().
2072 * Please notice that if the file is not released, the previous
2073 * MDS_DATA_MODIFIED has taken effect and usually
2074 * LLIF_DATA_MODIFIED is not set(see vvp_io_setattr_fini()).
2075 * This way we can save an RPC for common open + trunc
2077 if (ll_file_test_and_clear_flag(lli, LLIF_DATA_MODIFIED)) {
2078 struct hsm_state_set hss = {
2079 .hss_valid = HSS_SETMASK,
2080 .hss_setmask = HS_DIRTY,
2084 rc2 = ll_hsm_state_set(inode, &hss);
2085 /* truncate and write can happen at the same time, so that
2086 * the file can be set modified even though the file is not
2087 * restored from released state, and ll_hsm_state_set() is
2088 * not applicable for the file, and rc2 < 0 is normal in this
2091 CDEBUG(D_INFO, DFID "HSM set dirty failed: rc2 = %d\n",
2092 PFID(ll_inode2fid(inode)), rc2);
2097 if (op_data != NULL)
2098 ll_finish_md_op_data(op_data);
2100 if (S_ISREG(inode->i_mode)) {
2102 if ((attr->ia_valid & ATTR_SIZE) && !hsm_import)
2103 inode_dio_wait(inode);
2104 /* Once we've got the i_mutex, it's safe to set the S_NOSEC
2105 * flag. ll_update_inode (called from ll_md_setattr), clears
2106 * inode flags, so there is a gap where S_NOSEC is not set.
2107 * This can cause a writer to take the i_mutex unnecessarily,
2108 * but this is safe to do and should be rare. */
2109 inode_has_no_xattr(inode);
2113 ll_stats_ops_tally(ll_i2sbi(inode), attr->ia_valid & ATTR_SIZE ?
2114 LPROC_LL_TRUNC : LPROC_LL_SETATTR,
2115 ktime_us_delta(ktime_get(), kstart));
2120 int ll_setattr(struct dentry *de, struct iattr *attr)
2122 int mode = de->d_inode->i_mode;
2123 enum op_xvalid xvalid = 0;
2126 rc = llcrypt_prepare_setattr(de, attr);
2130 if ((attr->ia_valid & (ATTR_CTIME|ATTR_SIZE|ATTR_MODE)) ==
2131 (ATTR_CTIME|ATTR_SIZE|ATTR_MODE))
2132 xvalid |= OP_XVALID_OWNEROVERRIDE;
2134 if (((attr->ia_valid & (ATTR_MODE|ATTR_FORCE|ATTR_SIZE)) ==
2135 (ATTR_SIZE|ATTR_MODE)) &&
2136 (((mode & S_ISUID) && !(attr->ia_mode & S_ISUID)) ||
2137 (((mode & (S_ISGID|S_IXGRP)) == (S_ISGID|S_IXGRP)) &&
2138 !(attr->ia_mode & S_ISGID))))
2139 attr->ia_valid |= ATTR_FORCE;
2141 if ((attr->ia_valid & ATTR_MODE) &&
2143 !(attr->ia_mode & S_ISUID) &&
2144 !(attr->ia_valid & ATTR_KILL_SUID))
2145 attr->ia_valid |= ATTR_KILL_SUID;
2147 if ((attr->ia_valid & ATTR_MODE) &&
2148 ((mode & (S_ISGID|S_IXGRP)) == (S_ISGID|S_IXGRP)) &&
2149 !(attr->ia_mode & S_ISGID) &&
2150 !(attr->ia_valid & ATTR_KILL_SGID))
2151 attr->ia_valid |= ATTR_KILL_SGID;
2153 return ll_setattr_raw(de, attr, xvalid, false);
2156 int ll_statfs_internal(struct ll_sb_info *sbi, struct obd_statfs *osfs,
2159 struct obd_statfs obd_osfs = { 0 };
2164 max_age = ktime_get_seconds() - sbi->ll_statfs_max_age;
2166 if (sbi->ll_flags & LL_SBI_LAZYSTATFS)
2167 flags |= OBD_STATFS_NODELAY;
2169 rc = obd_statfs(NULL, sbi->ll_md_exp, osfs, max_age, flags);
2173 osfs->os_type = LL_SUPER_MAGIC;
2175 CDEBUG(D_SUPER, "MDC blocks %llu/%llu objects %llu/%llu\n",
2176 osfs->os_bavail, osfs->os_blocks, osfs->os_ffree, osfs->os_files);
2178 if (osfs->os_state & OS_STATFS_SUM)
2181 rc = obd_statfs(NULL, sbi->ll_dt_exp, &obd_osfs, max_age, flags);
2182 if (rc) /* Possibly a filesystem with no OSTs. Report MDT totals. */
2185 CDEBUG(D_SUPER, "OSC blocks %llu/%llu objects %llu/%llu\n",
2186 obd_osfs.os_bavail, obd_osfs.os_blocks, obd_osfs.os_ffree,
2189 osfs->os_bsize = obd_osfs.os_bsize;
2190 osfs->os_blocks = obd_osfs.os_blocks;
2191 osfs->os_bfree = obd_osfs.os_bfree;
2192 osfs->os_bavail = obd_osfs.os_bavail;
2194 /* If we have _some_ OSTs, but don't have as many free objects on the
2195 * OSTs as inodes on the MDTs, reduce the reported number of inodes
2196 * to compensate, so that the "inodes in use" number is correct.
2197 * This should be kept in sync with lod_statfs() behaviour.
2199 if (obd_osfs.os_files && obd_osfs.os_ffree < osfs->os_ffree) {
2200 osfs->os_files = (osfs->os_files - osfs->os_ffree) +
2202 osfs->os_ffree = obd_osfs.os_ffree;
2209 static int ll_statfs_project(struct inode *inode, struct kstatfs *sfs)
2211 struct if_quotactl qctl = {
2212 .qc_cmd = LUSTRE_Q_GETQUOTA,
2213 .qc_type = PRJQUOTA,
2214 .qc_valid = QC_GENERAL,
2216 u64 limit, curblock;
2219 qctl.qc_id = ll_i2info(inode)->lli_projid;
2220 ret = quotactl_ioctl(ll_i2sbi(inode), &qctl);
2222 /* ignore errors if project ID does not have
2223 * a quota limit or feature unsupported.
2225 if (ret == -ESRCH || ret == -EOPNOTSUPP)
2230 limit = ((qctl.qc_dqblk.dqb_bsoftlimit ?
2231 qctl.qc_dqblk.dqb_bsoftlimit :
2232 qctl.qc_dqblk.dqb_bhardlimit) * 1024) / sfs->f_bsize;
2233 if (limit && sfs->f_blocks > limit) {
2234 curblock = (qctl.qc_dqblk.dqb_curspace +
2235 sfs->f_bsize - 1) / sfs->f_bsize;
2236 sfs->f_blocks = limit;
2237 sfs->f_bfree = sfs->f_bavail =
2238 (sfs->f_blocks > curblock) ?
2239 (sfs->f_blocks - curblock) : 0;
2242 limit = qctl.qc_dqblk.dqb_isoftlimit ?
2243 qctl.qc_dqblk.dqb_isoftlimit :
2244 qctl.qc_dqblk.dqb_ihardlimit;
2245 if (limit && sfs->f_files > limit) {
2246 sfs->f_files = limit;
2247 sfs->f_ffree = (sfs->f_files >
2248 qctl.qc_dqblk.dqb_curinodes) ?
2249 (sfs->f_files - qctl.qc_dqblk.dqb_curinodes) : 0;
2255 int ll_statfs(struct dentry *de, struct kstatfs *sfs)
2257 struct super_block *sb = de->d_sb;
2258 struct obd_statfs osfs;
2259 __u64 fsid = huge_encode_dev(sb->s_dev);
2260 ktime_t kstart = ktime_get();
2263 CDEBUG(D_VFSTRACE, "VFS Op:sb=%s (%p)\n", sb->s_id, sb);
2265 /* Some amount of caching on the client is allowed */
2266 rc = ll_statfs_internal(ll_s2sbi(sb), &osfs, OBD_STATFS_SUM);
2270 statfs_unpack(sfs, &osfs);
2272 /* We need to downshift for all 32-bit kernels, because we can't
2273 * tell if the kernel is being called via sys_statfs64() or not.
2274 * Stop before overflowing f_bsize - in which case it is better
2275 * to just risk EOVERFLOW if caller is using old sys_statfs(). */
2276 if (sizeof(long) < 8) {
2277 while (osfs.os_blocks > ~0UL && sfs->f_bsize < 0x40000000) {
2280 osfs.os_blocks >>= 1;
2281 osfs.os_bfree >>= 1;
2282 osfs.os_bavail >>= 1;
2286 sfs->f_blocks = osfs.os_blocks;
2287 sfs->f_bfree = osfs.os_bfree;
2288 sfs->f_bavail = osfs.os_bavail;
2289 sfs->f_fsid.val[0] = (__u32)fsid;
2290 sfs->f_fsid.val[1] = (__u32)(fsid >> 32);
2291 if (ll_i2info(de->d_inode)->lli_projid)
2292 return ll_statfs_project(de->d_inode, sfs);
2294 ll_stats_ops_tally(ll_s2sbi(sb), LPROC_LL_STATFS,
2295 ktime_us_delta(ktime_get(), kstart));
2300 void ll_inode_size_lock(struct inode *inode)
2302 struct ll_inode_info *lli;
2304 LASSERT(!S_ISDIR(inode->i_mode));
2306 lli = ll_i2info(inode);
2307 mutex_lock(&lli->lli_size_mutex);
2310 void ll_inode_size_unlock(struct inode *inode)
2312 struct ll_inode_info *lli;
2314 lli = ll_i2info(inode);
2315 mutex_unlock(&lli->lli_size_mutex);
2318 void ll_update_inode_flags(struct inode *inode, int ext_flags)
2320 /* do not clear encryption flag */
2321 ext_flags |= ll_inode_to_ext_flags(inode->i_flags) & LUSTRE_ENCRYPT_FL;
2322 inode->i_flags = ll_ext_to_inode_flags(ext_flags);
2323 if (ext_flags & LUSTRE_PROJINHERIT_FL)
2324 ll_file_set_flag(ll_i2info(inode), LLIF_PROJECT_INHERIT);
2326 ll_file_clear_flag(ll_i2info(inode), LLIF_PROJECT_INHERIT);
2329 int ll_update_inode(struct inode *inode, struct lustre_md *md)
2331 struct ll_inode_info *lli = ll_i2info(inode);
2332 struct mdt_body *body = md->body;
2333 struct ll_sb_info *sbi = ll_i2sbi(inode);
2336 if (body->mbo_valid & OBD_MD_FLEASIZE) {
2337 rc = cl_file_inode_init(inode, md);
2342 if (S_ISDIR(inode->i_mode)) {
2343 rc = ll_update_lsm_md(inode, md);
2348 if (body->mbo_valid & OBD_MD_FLACL)
2349 lli_replace_acl(lli, md);
2351 inode->i_ino = cl_fid_build_ino(&body->mbo_fid1,
2352 sbi->ll_flags & LL_SBI_32BIT_API);
2353 inode->i_generation = cl_fid_build_gen(&body->mbo_fid1);
2355 if (body->mbo_valid & OBD_MD_FLATIME) {
2356 if (body->mbo_atime > inode->i_atime.tv_sec)
2357 inode->i_atime.tv_sec = body->mbo_atime;
2358 lli->lli_atime = body->mbo_atime;
2361 if (body->mbo_valid & OBD_MD_FLMTIME) {
2362 if (body->mbo_mtime > inode->i_mtime.tv_sec) {
2364 "setting ino %lu mtime from %lld to %llu\n",
2365 inode->i_ino, (s64)inode->i_mtime.tv_sec,
2367 inode->i_mtime.tv_sec = body->mbo_mtime;
2369 lli->lli_mtime = body->mbo_mtime;
2372 if (body->mbo_valid & OBD_MD_FLCTIME) {
2373 if (body->mbo_ctime > inode->i_ctime.tv_sec)
2374 inode->i_ctime.tv_sec = body->mbo_ctime;
2375 lli->lli_ctime = body->mbo_ctime;
2378 if (body->mbo_valid & OBD_MD_FLBTIME)
2379 lli->lli_btime = body->mbo_btime;
2381 /* Clear i_flags to remove S_NOSEC before permissions are updated */
2382 if (body->mbo_valid & OBD_MD_FLFLAGS)
2383 ll_update_inode_flags(inode, body->mbo_flags);
2384 if (body->mbo_valid & OBD_MD_FLMODE)
2385 inode->i_mode = (inode->i_mode & S_IFMT) |
2386 (body->mbo_mode & ~S_IFMT);
2388 if (body->mbo_valid & OBD_MD_FLTYPE)
2389 inode->i_mode = (inode->i_mode & ~S_IFMT) |
2390 (body->mbo_mode & S_IFMT);
2392 LASSERT(inode->i_mode != 0);
2393 if (body->mbo_valid & OBD_MD_FLUID)
2394 inode->i_uid = make_kuid(&init_user_ns, body->mbo_uid);
2395 if (body->mbo_valid & OBD_MD_FLGID)
2396 inode->i_gid = make_kgid(&init_user_ns, body->mbo_gid);
2397 if (body->mbo_valid & OBD_MD_FLPROJID)
2398 lli->lli_projid = body->mbo_projid;
2399 if (body->mbo_valid & OBD_MD_FLNLINK)
2400 set_nlink(inode, body->mbo_nlink);
2401 if (body->mbo_valid & OBD_MD_FLRDEV)
2402 inode->i_rdev = old_decode_dev(body->mbo_rdev);
2404 if (body->mbo_valid & OBD_MD_FLID) {
2405 /* FID shouldn't be changed! */
2406 if (fid_is_sane(&lli->lli_fid)) {
2407 LASSERTF(lu_fid_eq(&lli->lli_fid, &body->mbo_fid1),
2408 "Trying to change FID "DFID
2409 " to the "DFID", inode "DFID"(%p)\n",
2410 PFID(&lli->lli_fid), PFID(&body->mbo_fid1),
2411 PFID(ll_inode2fid(inode)), inode);
2413 lli->lli_fid = body->mbo_fid1;
2417 LASSERT(fid_seq(&lli->lli_fid) != 0);
2419 lli->lli_attr_valid = body->mbo_valid;
2420 if (body->mbo_valid & OBD_MD_FLSIZE) {
2421 i_size_write(inode, body->mbo_size);
2423 CDEBUG(D_VFSTRACE, "inode="DFID", updating i_size %llu\n",
2424 PFID(ll_inode2fid(inode)),
2425 (unsigned long long)body->mbo_size);
2427 if (body->mbo_valid & OBD_MD_FLBLOCKS)
2428 inode->i_blocks = body->mbo_blocks;
2430 if (body->mbo_valid & OBD_MD_FLLAZYSIZE)
2431 lli->lli_lazysize = body->mbo_size;
2432 if (body->mbo_valid & OBD_MD_FLLAZYBLOCKS)
2433 lli->lli_lazyblocks = body->mbo_blocks;
2436 if (body->mbo_valid & OBD_MD_TSTATE) {
2437 /* Set LLIF_FILE_RESTORING if restore ongoing and
2438 * clear it when done to ensure to start again
2439 * glimpsing updated attrs
2441 if (body->mbo_t_state & MS_RESTORE)
2442 ll_file_set_flag(lli, LLIF_FILE_RESTORING);
2444 ll_file_clear_flag(lli, LLIF_FILE_RESTORING);
2450 int ll_read_inode2(struct inode *inode, void *opaque)
2452 struct lustre_md *md = opaque;
2453 struct ll_inode_info *lli = ll_i2info(inode);
2457 CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p)\n",
2458 PFID(&lli->lli_fid), inode);
2460 /* Core attributes from the MDS first. This is a new inode, and
2461 * the VFS doesn't zero times in the core inode so we have to do
2462 * it ourselves. They will be overwritten by either MDS or OST
2463 * attributes - we just need to make sure they aren't newer.
2465 inode->i_mtime.tv_sec = 0;
2466 inode->i_atime.tv_sec = 0;
2467 inode->i_ctime.tv_sec = 0;
2469 rc = ll_update_inode(inode, md);
2473 /* OIDEBUG(inode); */
2475 #ifdef HAVE_BACKING_DEV_INFO
2476 /* initializing backing dev info. */
2477 inode->i_mapping->backing_dev_info = &s2lsi(inode->i_sb)->lsi_bdi;
2479 if (S_ISREG(inode->i_mode)) {
2480 struct ll_sb_info *sbi = ll_i2sbi(inode);
2481 inode->i_op = &ll_file_inode_operations;
2482 inode->i_fop = sbi->ll_fop;
2483 inode->i_mapping->a_ops = (struct address_space_operations *)&ll_aops;
2485 } else if (S_ISDIR(inode->i_mode)) {
2486 inode->i_op = &ll_dir_inode_operations;
2487 inode->i_fop = &ll_dir_operations;
2489 } else if (S_ISLNK(inode->i_mode)) {
2490 inode->i_op = &ll_fast_symlink_inode_operations;
2493 inode->i_op = &ll_special_inode_operations;
2495 init_special_inode(inode, inode->i_mode,
2504 void ll_delete_inode(struct inode *inode)
2506 struct ll_inode_info *lli = ll_i2info(inode);
2507 struct address_space *mapping = &inode->i_data;
2508 unsigned long nrpages;
2509 unsigned long flags;
2513 if (S_ISREG(inode->i_mode) && lli->lli_clob != NULL) {
2514 /* It is last chance to write out dirty pages,
2515 * otherwise we may lose data while umount.
2517 * If i_nlink is 0 then just discard data. This is safe because
2518 * local inode gets i_nlink 0 from server only for the last
2519 * unlink, so that file is not opened somewhere else
2521 cl_sync_file_range(inode, 0, OBD_OBJECT_EOF, inode->i_nlink ?
2522 CL_FSYNC_LOCAL : CL_FSYNC_DISCARD, 1);
2524 truncate_inode_pages_final(mapping);
2526 /* Workaround for LU-118: Note nrpages may not be totally updated when
2527 * truncate_inode_pages() returns, as there can be a page in the process
2528 * of deletion (inside __delete_from_page_cache()) in the specified
2529 * range. Thus mapping->nrpages can be non-zero when this function
2530 * returns even after truncation of the whole mapping. Only do this if
2531 * npages isn't already zero.
2533 nrpages = mapping->nrpages;
2535 ll_xa_lock_irqsave(&mapping->i_pages, flags);
2536 nrpages = mapping->nrpages;
2537 ll_xa_unlock_irqrestore(&mapping->i_pages, flags);
2538 } /* Workaround end */
2540 LASSERTF(nrpages == 0, "%s: inode="DFID"(%p) nrpages=%lu, "
2541 "see https://jira.whamcloud.com/browse/LU-118\n",
2542 ll_i2sbi(inode)->ll_fsname,
2543 PFID(ll_inode2fid(inode)), inode, nrpages);
2545 ll_clear_inode(inode);
2551 int ll_iocontrol(struct inode *inode, struct file *file,
2552 unsigned int cmd, unsigned long arg)
2554 struct ll_sb_info *sbi = ll_i2sbi(inode);
2555 struct ptlrpc_request *req = NULL;
2560 case FS_IOC_GETFLAGS: {
2561 struct mdt_body *body;
2562 struct md_op_data *op_data;
2564 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL,
2565 0, 0, LUSTRE_OPC_ANY,
2567 if (IS_ERR(op_data))
2568 RETURN(PTR_ERR(op_data));
2570 op_data->op_valid = OBD_MD_FLFLAGS;
2571 rc = md_getattr(sbi->ll_md_exp, op_data, &req);
2572 ll_finish_md_op_data(op_data);
2574 CERROR("%s: failure inode "DFID": rc = %d\n",
2575 sbi->ll_md_exp->exp_obd->obd_name,
2576 PFID(ll_inode2fid(inode)), rc);
2580 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
2582 flags = body->mbo_flags;
2584 ptlrpc_req_finished(req);
2586 RETURN(put_user(flags, (int __user *)arg));
2588 case FS_IOC_SETFLAGS: {
2590 struct md_op_data *op_data;
2591 struct cl_object *obj;
2592 struct fsxattr fa = { 0 };
2594 if (get_user(flags, (int __user *)arg))
2597 fa.fsx_projid = ll_i2info(inode)->lli_projid;
2598 if (flags & LUSTRE_PROJINHERIT_FL)
2599 fa.fsx_xflags = FS_XFLAG_PROJINHERIT;
2601 rc = ll_ioctl_check_project(inode, &fa);
2605 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
2606 LUSTRE_OPC_ANY, NULL);
2607 if (IS_ERR(op_data))
2608 RETURN(PTR_ERR(op_data));
2610 op_data->op_attr_flags = flags;
2611 op_data->op_xvalid |= OP_XVALID_FLAGS;
2612 rc = md_setattr(sbi->ll_md_exp, op_data, NULL, 0, &req);
2613 ll_finish_md_op_data(op_data);
2614 ptlrpc_req_finished(req);
2618 ll_update_inode_flags(inode, flags);
2620 obj = ll_i2info(inode)->lli_clob;
2624 OBD_ALLOC_PTR(attr);
2628 rc = cl_setattr_ost(obj, attr, OP_XVALID_FLAGS, flags);
2640 int ll_flush_ctx(struct inode *inode)
2642 struct ll_sb_info *sbi = ll_i2sbi(inode);
2644 CDEBUG(D_SEC, "flush context for user %d\n",
2645 from_kuid(&init_user_ns, current_uid()));
2647 obd_set_info_async(NULL, sbi->ll_md_exp,
2648 sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
2650 obd_set_info_async(NULL, sbi->ll_dt_exp,
2651 sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
2656 /* umount -f client means force down, don't save state */
2657 void ll_umount_begin(struct super_block *sb)
2659 struct ll_sb_info *sbi = ll_s2sbi(sb);
2660 struct obd_device *obd;
2661 struct obd_ioctl_data *ioc_data;
2665 CDEBUG(D_VFSTRACE, "VFS Op: superblock %p count %d active %d\n", sb,
2666 sb->s_count, atomic_read(&sb->s_active));
2668 obd = class_exp2obd(sbi->ll_md_exp);
2670 CERROR("Invalid MDC connection handle %#llx\n",
2671 sbi->ll_md_exp->exp_handle.h_cookie);
2677 obd = class_exp2obd(sbi->ll_dt_exp);
2679 CERROR("Invalid LOV connection handle %#llx\n",
2680 sbi->ll_dt_exp->exp_handle.h_cookie);
2686 OBD_ALLOC_PTR(ioc_data);
2688 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_md_exp,
2689 sizeof *ioc_data, ioc_data, NULL);
2691 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_dt_exp,
2692 sizeof *ioc_data, ioc_data, NULL);
2694 OBD_FREE_PTR(ioc_data);
2697 /* Really, we'd like to wait until there are no requests outstanding,
2698 * and then continue. For now, we just periodically checking for vfs
2699 * to decrement mnt_cnt and hope to finish it within 10sec.
2703 !may_umount(sbi->ll_mnt.mnt)) {
2711 int ll_remount_fs(struct super_block *sb, int *flags, char *data)
2713 struct ll_sb_info *sbi = ll_s2sbi(sb);
2714 char *profilenm = get_profile_name(sb);
2718 if ((*flags & MS_RDONLY) != (sb->s_flags & SB_RDONLY)) {
2719 read_only = *flags & MS_RDONLY;
2720 err = obd_set_info_async(NULL, sbi->ll_md_exp,
2721 sizeof(KEY_READ_ONLY),
2722 KEY_READ_ONLY, sizeof(read_only),
2725 LCONSOLE_WARN("Failed to remount %s %s (%d)\n",
2726 profilenm, read_only ?
2727 "read-only" : "read-write", err);
2732 sb->s_flags |= SB_RDONLY;
2734 sb->s_flags &= ~SB_RDONLY;
2736 if (sbi->ll_flags & LL_SBI_VERBOSE)
2737 LCONSOLE_WARN("Remounted %s %s\n", profilenm,
2738 read_only ? "read-only" : "read-write");
2744 * Cleanup the open handle that is cached on MDT-side.
2746 * For open case, the client side open handling thread may hit error
2747 * after the MDT grant the open. Under such case, the client should
2748 * send close RPC to the MDT as cleanup; otherwise, the open handle
2749 * on the MDT will be leaked there until the client umount or evicted.
2751 * In further, if someone unlinked the file, because the open handle
2752 * holds the reference on such file/object, then it will block the
2753 * subsequent threads that want to locate such object via FID.
2755 * \param[in] sb super block for this file-system
2756 * \param[in] open_req pointer to the original open request
2758 void ll_open_cleanup(struct super_block *sb, struct ptlrpc_request *open_req)
2760 struct mdt_body *body;
2761 struct md_op_data *op_data;
2762 struct ptlrpc_request *close_req = NULL;
2763 struct obd_export *exp = ll_s2sbi(sb)->ll_md_exp;
2766 body = req_capsule_server_get(&open_req->rq_pill, &RMF_MDT_BODY);
2767 OBD_ALLOC_PTR(op_data);
2768 if (op_data == NULL) {
2769 CWARN("%s: cannot allocate op_data to release open handle for "
2770 DFID"\n", ll_s2sbi(sb)->ll_fsname, PFID(&body->mbo_fid1));
2775 op_data->op_fid1 = body->mbo_fid1;
2776 op_data->op_open_handle = body->mbo_open_handle;
2777 op_data->op_mod_time = ktime_get_real_seconds();
2778 md_close(exp, op_data, NULL, &close_req);
2779 ptlrpc_req_finished(close_req);
2780 ll_finish_md_op_data(op_data);
2785 int ll_prep_inode(struct inode **inode, struct ptlrpc_request *req,
2786 struct super_block *sb, struct lookup_intent *it)
2788 struct ll_sb_info *sbi = NULL;
2789 struct lustre_md md = { NULL };
2790 bool default_lmv_deleted = false;
2795 LASSERT(*inode || sb);
2796 sbi = sb ? ll_s2sbi(sb) : ll_i2sbi(*inode);
2797 rc = md_get_lustre_md(sbi->ll_md_exp, req, sbi->ll_dt_exp,
2798 sbi->ll_md_exp, &md);
2803 * clear default_lmv only if intent_getattr reply doesn't contain it.
2804 * but it needs to be done after iget, check this early because
2805 * ll_update_lsm_md() may change md.
2807 if (it && (it->it_op & (IT_LOOKUP | IT_GETATTR)) &&
2808 S_ISDIR(md.body->mbo_mode) && !md.default_lmv)
2809 default_lmv_deleted = true;
2812 rc = ll_update_inode(*inode, &md);
2816 LASSERT(sb != NULL);
2819 * At this point server returns to client's same fid as client
2820 * generated for creating. So using ->fid1 is okay here.
2822 if (!fid_is_sane(&md.body->mbo_fid1)) {
2823 CERROR("%s: Fid is insane "DFID"\n",
2825 PFID(&md.body->mbo_fid1));
2826 GOTO(out, rc = -EINVAL);
2829 *inode = ll_iget(sb, cl_fid_build_ino(&md.body->mbo_fid1,
2830 sbi->ll_flags & LL_SBI_32BIT_API),
2832 if (IS_ERR(*inode)) {
2834 rc = IS_ERR(*inode) ? PTR_ERR(*inode) : -ENOMEM;
2836 CERROR("new_inode -fatal: rc %d\n", rc);
2841 /* Handling piggyback layout lock.
2842 * Layout lock can be piggybacked by getattr and open request.
2843 * The lsm can be applied to inode only if it comes with a layout lock
2844 * otherwise correct layout may be overwritten, for example:
2845 * 1. proc1: mdt returns a lsm but not granting layout
2846 * 2. layout was changed by another client
2847 * 3. proc2: refresh layout and layout lock granted
2848 * 4. proc1: to apply a stale layout */
2849 if (it != NULL && it->it_lock_mode != 0) {
2850 struct lustre_handle lockh;
2851 struct ldlm_lock *lock;
2853 lockh.cookie = it->it_lock_handle;
2854 lock = ldlm_handle2lock(&lockh);
2855 LASSERT(lock != NULL);
2856 if (ldlm_has_layout(lock)) {
2857 struct cl_object_conf conf;
2859 memset(&conf, 0, sizeof(conf));
2860 conf.coc_opc = OBJECT_CONF_SET;
2861 conf.coc_inode = *inode;
2862 conf.coc_lock = lock;
2863 conf.u.coc_layout = md.layout;
2864 (void)ll_layout_conf(*inode, &conf);
2866 LDLM_LOCK_PUT(lock);
2869 if (default_lmv_deleted)
2870 ll_update_default_lsm_md(*inode, &md);
2875 /* cleanup will be done if necessary */
2876 md_free_lustre_md(sbi->ll_md_exp, &md);
2878 if (rc != 0 && it != NULL && it->it_op & IT_OPEN) {
2879 ll_intent_drop_lock(it);
2880 ll_open_cleanup(sb != NULL ? sb : (*inode)->i_sb, req);
2886 int ll_obd_statfs(struct inode *inode, void __user *arg)
2888 struct ll_sb_info *sbi = NULL;
2889 struct obd_export *exp;
2891 struct obd_ioctl_data *data = NULL;
2895 if (!inode || !(sbi = ll_i2sbi(inode)))
2896 GOTO(out_statfs, rc = -EINVAL);
2898 rc = obd_ioctl_getdata(&buf, &len, arg);
2900 GOTO(out_statfs, rc);
2903 if (!data->ioc_inlbuf1 || !data->ioc_inlbuf2 ||
2904 !data->ioc_pbuf1 || !data->ioc_pbuf2)
2905 GOTO(out_statfs, rc = -EINVAL);
2907 if (data->ioc_inllen1 != sizeof(__u32) ||
2908 data->ioc_inllen2 != sizeof(__u32) ||
2909 data->ioc_plen1 != sizeof(struct obd_statfs) ||
2910 data->ioc_plen2 != sizeof(struct obd_uuid))
2911 GOTO(out_statfs, rc = -EINVAL);
2913 memcpy(&type, data->ioc_inlbuf1, sizeof(__u32));
2914 if (type & LL_STATFS_LMV)
2915 exp = sbi->ll_md_exp;
2916 else if (type & LL_STATFS_LOV)
2917 exp = sbi->ll_dt_exp;
2919 GOTO(out_statfs, rc = -ENODEV);
2921 rc = obd_iocontrol(IOC_OBD_STATFS, exp, len, buf, NULL);
2923 GOTO(out_statfs, rc);
2925 OBD_FREE_LARGE(buf, len);
2930 * this is normally called in ll_fini_md_op_data(), but sometimes it needs to
2931 * be called early to avoid deadlock.
2933 void ll_unlock_md_op_lsm(struct md_op_data *op_data)
2935 if (op_data->op_mea2_sem) {
2936 up_read_non_owner(op_data->op_mea2_sem);
2937 op_data->op_mea2_sem = NULL;
2940 if (op_data->op_mea1_sem) {
2941 up_read_non_owner(op_data->op_mea1_sem);
2942 op_data->op_mea1_sem = NULL;
2946 /* this function prepares md_op_data hint for passing it down to MD stack. */
2947 struct md_op_data *ll_prep_md_op_data(struct md_op_data *op_data,
2948 struct inode *i1, struct inode *i2,
2949 const char *name, size_t namelen,
2950 __u32 mode, enum md_op_code opc,
2953 LASSERT(i1 != NULL);
2956 /* Do not reuse namelen for something else. */
2958 return ERR_PTR(-EINVAL);
2960 if (namelen > ll_i2sbi(i1)->ll_namelen)
2961 return ERR_PTR(-ENAMETOOLONG);
2963 /* "/" is not valid name, but it's allowed */
2964 if (!lu_name_is_valid_2(name, namelen) &&
2965 strncmp("/", name, namelen) != 0)
2966 return ERR_PTR(-EINVAL);
2969 if (op_data == NULL)
2970 OBD_ALLOC_PTR(op_data);
2972 if (op_data == NULL)
2973 return ERR_PTR(-ENOMEM);
2975 ll_i2gids(op_data->op_suppgids, i1, i2);
2976 op_data->op_fid1 = *ll_inode2fid(i1);
2977 op_data->op_code = opc;
2979 if (S_ISDIR(i1->i_mode)) {
2980 down_read_non_owner(&ll_i2info(i1)->lli_lsm_sem);
2981 op_data->op_mea1_sem = &ll_i2info(i1)->lli_lsm_sem;
2982 op_data->op_mea1 = ll_i2info(i1)->lli_lsm_md;
2983 op_data->op_default_mea1 = ll_i2info(i1)->lli_default_lsm_md;
2987 op_data->op_fid2 = *ll_inode2fid(i2);
2988 if (S_ISDIR(i2->i_mode)) {
2990 /* i2 is typically a child of i1, and MUST be
2991 * further from the root to avoid deadlocks.
2993 down_read_non_owner(&ll_i2info(i2)->lli_lsm_sem);
2994 op_data->op_mea2_sem =
2995 &ll_i2info(i2)->lli_lsm_sem;
2997 op_data->op_mea2 = ll_i2info(i2)->lli_lsm_md;
3000 fid_zero(&op_data->op_fid2);
3003 if (ll_i2sbi(i1)->ll_flags & LL_SBI_64BIT_HASH)
3004 op_data->op_cli_flags |= CLI_HASH64;
3006 if (ll_need_32bit_api(ll_i2sbi(i1)))
3007 op_data->op_cli_flags |= CLI_API32;
3009 op_data->op_name = name;
3010 op_data->op_namelen = namelen;
3011 op_data->op_mode = mode;
3012 op_data->op_mod_time = ktime_get_real_seconds();
3013 op_data->op_fsuid = from_kuid(&init_user_ns, current_fsuid());
3014 op_data->op_fsgid = from_kgid(&init_user_ns, current_fsgid());
3015 op_data->op_cap = cfs_curproc_cap_pack();
3016 op_data->op_mds = 0;
3017 if ((opc == LUSTRE_OPC_CREATE) && (name != NULL) &&
3018 filename_is_volatile(name, namelen, &op_data->op_mds)) {
3019 op_data->op_bias |= MDS_CREATE_VOLATILE;
3021 op_data->op_data = data;
3026 void ll_finish_md_op_data(struct md_op_data *op_data)
3028 ll_unlock_md_op_lsm(op_data);
3029 security_release_secctx(op_data->op_file_secctx,
3030 op_data->op_file_secctx_size);
3031 llcrypt_free_ctx(op_data->op_file_encctx, op_data->op_file_encctx_size);
3032 OBD_FREE_PTR(op_data);
3035 int ll_show_options(struct seq_file *seq, struct dentry *dentry)
3037 struct ll_sb_info *sbi;
3039 LASSERT(seq && dentry);
3040 sbi = ll_s2sbi(dentry->d_sb);
3042 if (sbi->ll_flags & LL_SBI_NOLCK)
3043 seq_puts(seq, ",nolock");
3045 /* "flock" is the default since 2.13, but it wasn't for many years,
3046 * so it is still useful to print this to show it is enabled.
3047 * Start to print "noflock" so it is now clear when flock is disabled.
3049 if (sbi->ll_flags & LL_SBI_FLOCK)
3050 seq_puts(seq, ",flock");
3051 else if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
3052 seq_puts(seq, ",localflock");
3054 seq_puts(seq, ",noflock");
3056 if (sbi->ll_flags & LL_SBI_USER_XATTR)
3057 seq_puts(seq, ",user_xattr");
3059 if (sbi->ll_flags & LL_SBI_LAZYSTATFS)
3060 seq_puts(seq, ",lazystatfs");
3062 if (sbi->ll_flags & LL_SBI_USER_FID2PATH)
3063 seq_puts(seq, ",user_fid2path");
3065 if (sbi->ll_flags & LL_SBI_ALWAYS_PING)
3066 seq_puts(seq, ",always_ping");
3068 if (ll_sbi_has_test_dummy_encryption(sbi))
3069 seq_puts(seq, ",test_dummy_encryption");
3071 if (ll_sbi_has_encrypt(sbi))
3072 seq_puts(seq, ",encrypt");
3074 seq_puts(seq, ",noencrypt");
3080 * Get obd name by cmd, and copy out to user space
3082 int ll_get_obd_name(struct inode *inode, unsigned int cmd, unsigned long arg)
3084 struct ll_sb_info *sbi = ll_i2sbi(inode);
3085 struct obd_device *obd;
3088 if (cmd == OBD_IOC_GETDTNAME)
3089 obd = class_exp2obd(sbi->ll_dt_exp);
3090 else if (cmd == OBD_IOC_GETMDNAME)
3091 obd = class_exp2obd(sbi->ll_md_exp);
3098 if (copy_to_user((void __user *)arg, obd->obd_name,
3099 strlen(obd->obd_name) + 1))
3105 static char* ll_d_path(struct dentry *dentry, char *buf, int bufsize)
3112 p.mnt = current->fs->root.mnt;
3114 path = d_path(&p, buf, bufsize);
3119 void ll_dirty_page_discard_warn(struct page *page, int ioret)
3121 char *buf, *path = NULL;
3122 struct dentry *dentry = NULL;
3123 struct inode *inode = page->mapping->host;
3125 /* this can be called inside spin lock so use GFP_ATOMIC. */
3126 buf = (char *)__get_free_page(GFP_ATOMIC);
3128 dentry = d_find_alias(page->mapping->host);
3130 path = ll_d_path(dentry, buf, PAGE_SIZE);
3133 /* The below message is checked in recovery-small.sh test_24b */
3135 "%s: dirty page discard: %s/fid: "DFID"/%s may get corrupted "
3136 "(rc %d)\n", ll_i2sbi(inode)->ll_fsname,
3137 s2lsi(page->mapping->host->i_sb)->lsi_lmd->lmd_dev,
3138 PFID(ll_inode2fid(inode)),
3139 (path && !IS_ERR(path)) ? path : "", ioret);
3145 free_page((unsigned long)buf);
3148 ssize_t ll_copy_user_md(const struct lov_user_md __user *md,
3149 struct lov_user_md **kbuf)
3151 struct lov_user_md lum;
3155 if (copy_from_user(&lum, md, sizeof(lum)))
3158 lum_size = ll_lov_user_md_size(&lum);
3162 OBD_ALLOC_LARGE(*kbuf, lum_size);
3166 if (copy_from_user(*kbuf, md, lum_size) != 0) {
3167 OBD_FREE_LARGE(*kbuf, lum_size);
3175 * Compute llite root squash state after a change of root squash
3176 * configuration setting or add/remove of a lnet nid
3178 void ll_compute_rootsquash_state(struct ll_sb_info *sbi)
3180 struct root_squash_info *squash = &sbi->ll_squash;
3183 struct lnet_process_id id;
3185 /* Update norootsquash flag */
3186 spin_lock(&squash->rsi_lock);
3187 if (list_empty(&squash->rsi_nosquash_nids))
3188 sbi->ll_flags &= ~LL_SBI_NOROOTSQUASH;
3190 /* Do not apply root squash as soon as one of our NIDs is
3191 * in the nosquash_nids list */
3194 while (LNetGetId(i++, &id) != -ENOENT) {
3195 if (id.nid == LNET_NID_LO_0)
3197 if (cfs_match_nid(id.nid, &squash->rsi_nosquash_nids)) {
3203 sbi->ll_flags |= LL_SBI_NOROOTSQUASH;
3205 sbi->ll_flags &= ~LL_SBI_NOROOTSQUASH;
3207 spin_unlock(&squash->rsi_lock);
3211 * Parse linkea content to extract information about a given hardlink
3213 * \param[in] ldata - Initialized linkea data
3214 * \param[in] linkno - Link identifier
3215 * \param[out] parent_fid - The entry's parent FID
3216 * \param[out] ln - Entry name destination buffer
3218 * \retval 0 on success
3219 * \retval Appropriate negative error code on failure
3221 static int ll_linkea_decode(struct linkea_data *ldata, unsigned int linkno,
3222 struct lu_fid *parent_fid, struct lu_name *ln)
3228 rc = linkea_init_with_rec(ldata);
3232 if (linkno >= ldata->ld_leh->leh_reccount)
3233 /* beyond last link */
3236 linkea_first_entry(ldata);
3237 for (idx = 0; ldata->ld_lee != NULL; idx++) {
3238 linkea_entry_unpack(ldata->ld_lee, &ldata->ld_reclen, ln,
3243 linkea_next_entry(ldata);
3253 * Get parent FID and name of an identified link. Operation is performed for
3254 * a given link number, letting the caller iterate over linkno to list one or
3255 * all links of an entry.
3257 * \param[in] file - File descriptor against which to perform the operation
3258 * \param[in,out] arg - User-filled structure containing the linkno to operate
3259 * on and the available size. It is eventually filled with
3260 * the requested information or left untouched on error
3262 * \retval - 0 on success
3263 * \retval - Appropriate negative error code on failure
3265 int ll_getparent(struct file *file, struct getparent __user *arg)
3267 struct inode *inode = file_inode(file);
3268 struct linkea_data *ldata;
3269 struct lu_buf buf = LU_BUF_NULL;
3271 struct lu_fid parent_fid;
3278 if (!cfs_capable(CFS_CAP_DAC_READ_SEARCH) &&
3279 !(ll_i2sbi(inode)->ll_flags & LL_SBI_USER_FID2PATH))
3282 if (get_user(name_size, &arg->gp_name_size))
3285 if (get_user(linkno, &arg->gp_linkno))
3288 if (name_size > PATH_MAX)
3291 OBD_ALLOC(ldata, sizeof(*ldata));
3295 rc = linkea_data_new(ldata, &buf);
3297 GOTO(ldata_free, rc);
3299 rc = ll_xattr_list(inode, XATTR_NAME_LINK, XATTR_TRUSTED_T, buf.lb_buf,
3300 buf.lb_len, OBD_MD_FLXATTR);
3304 rc = ll_linkea_decode(ldata, linkno, &parent_fid, &ln);
3308 if (ln.ln_namelen >= name_size)
3309 GOTO(lb_free, rc = -EOVERFLOW);
3311 if (copy_to_user(&arg->gp_fid, &parent_fid, sizeof(arg->gp_fid)))
3312 GOTO(lb_free, rc = -EFAULT);
3314 if (copy_to_user(&arg->gp_name, ln.ln_name, ln.ln_namelen))
3315 GOTO(lb_free, rc = -EFAULT);
3317 if (put_user('\0', arg->gp_name + ln.ln_namelen))
3318 GOTO(lb_free, rc = -EFAULT);
3323 OBD_FREE(ldata, sizeof(*ldata));