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/module.h>
40 #include <linux/random.h>
41 #include <linux/statfs.h>
42 #include <linux/time.h>
43 #include <linux/types.h>
44 #include <libcfs/linux/linux-uuid.h>
45 #include <linux/version.h>
47 #include <linux/user_namespace.h>
48 #ifdef HAVE_UIDGID_HEADER
49 # include <linux/uidgid.h>
51 #include <linux/security.h>
53 #include <uapi/linux/lustre/lustre_ioctl.h>
54 #ifdef HAVE_UAPI_LINUX_MOUNT_H
55 #include <uapi/linux/mount.h>
58 #include <lustre_ha.h>
59 #include <lustre_dlm.h>
60 #include <lprocfs_status.h>
61 #include <lustre_disk.h>
62 #include <uapi/linux/lustre/lustre_param.h>
63 #include <lustre_log.h>
64 #include <cl_object.h>
65 #include <obd_cksum.h>
66 #include "llite_internal.h"
68 struct kmem_cache *ll_file_data_slab;
71 #define log2(n) ffz(~(n))
74 static struct ll_sb_info *ll_init_sbi(void)
76 struct ll_sb_info *sbi = NULL;
78 unsigned long lru_page_max;
87 RETURN(ERR_PTR(-ENOMEM));
89 rc = pcc_super_init(&sbi->ll_pcc_super);
93 spin_lock_init(&sbi->ll_lock);
94 mutex_init(&sbi->ll_lco.lco_lock);
95 spin_lock_init(&sbi->ll_pp_extent_lock);
96 spin_lock_init(&sbi->ll_process_lock);
97 sbi->ll_rw_stats_on = 0;
98 sbi->ll_statfs_max_age = OBD_STATFS_CACHE_SECONDS;
101 pages = si.totalram - si.totalhigh;
102 lru_page_max = pages / 2;
104 sbi->ll_ra_info.ra_async_max_active = 0;
105 sbi->ll_ra_info.ll_readahead_wq =
106 alloc_workqueue("ll-readahead-wq", WQ_UNBOUND,
107 sbi->ll_ra_info.ra_async_max_active);
108 if (!sbi->ll_ra_info.ll_readahead_wq)
109 GOTO(out_pcc, rc = -ENOMEM);
111 /* initialize ll_cache data */
112 sbi->ll_cache = cl_cache_init(lru_page_max);
113 if (sbi->ll_cache == NULL)
114 GOTO(out_destroy_ra, rc = -ENOMEM);
116 sbi->ll_ra_info.ra_max_pages_per_file = min(pages / 32,
117 SBI_DEFAULT_READAHEAD_MAX);
118 sbi->ll_ra_info.ra_async_pages_per_file_threshold =
119 sbi->ll_ra_info.ra_max_pages_per_file;
120 sbi->ll_ra_info.ra_max_pages = sbi->ll_ra_info.ra_max_pages_per_file;
121 sbi->ll_ra_info.ra_max_read_ahead_whole_pages = -1;
123 sbi->ll_flags |= LL_SBI_VERBOSE;
124 #ifdef ENABLE_CHECKSUM
125 sbi->ll_flags |= LL_SBI_CHECKSUM;
128 sbi->ll_flags |= LL_SBI_FLOCK;
131 #ifdef HAVE_LRU_RESIZE_SUPPORT
132 sbi->ll_flags |= LL_SBI_LRU_RESIZE;
134 sbi->ll_flags |= LL_SBI_LAZYSTATFS;
136 for (i = 0; i <= LL_PROCESS_HIST_MAX; i++) {
137 spin_lock_init(&sbi->ll_rw_extents_info.pp_extents[i].
139 spin_lock_init(&sbi->ll_rw_extents_info.pp_extents[i].
143 /* metadata statahead is enabled by default */
144 sbi->ll_sa_running_max = LL_SA_RUNNING_DEF;
145 sbi->ll_sa_max = LL_SA_RPC_DEF;
146 atomic_set(&sbi->ll_sa_total, 0);
147 atomic_set(&sbi->ll_sa_wrong, 0);
148 atomic_set(&sbi->ll_sa_running, 0);
149 atomic_set(&sbi->ll_agl_total, 0);
150 sbi->ll_flags |= LL_SBI_AGL_ENABLED;
151 sbi->ll_flags |= LL_SBI_FAST_READ;
152 sbi->ll_flags |= LL_SBI_TINY_WRITE;
155 sbi->ll_squash.rsi_uid = 0;
156 sbi->ll_squash.rsi_gid = 0;
157 INIT_LIST_HEAD(&sbi->ll_squash.rsi_nosquash_nids);
158 spin_lock_init(&sbi->ll_squash.rsi_lock);
160 /* Per-filesystem file heat */
161 sbi->ll_heat_decay_weight = SBI_DEFAULT_HEAT_DECAY_WEIGHT;
162 sbi->ll_heat_period_second = SBI_DEFAULT_HEAT_PERIOD_SECOND;
165 destroy_workqueue(sbi->ll_ra_info.ll_readahead_wq);
167 pcc_super_fini(&sbi->ll_pcc_super);
173 static void ll_free_sbi(struct super_block *sb)
175 struct ll_sb_info *sbi = ll_s2sbi(sb);
179 if (!list_empty(&sbi->ll_squash.rsi_nosquash_nids))
180 cfs_free_nidlist(&sbi->ll_squash.rsi_nosquash_nids);
181 if (sbi->ll_ra_info.ll_readahead_wq)
182 destroy_workqueue(sbi->ll_ra_info.ll_readahead_wq);
183 if (sbi->ll_cache != NULL) {
184 cl_cache_decref(sbi->ll_cache);
185 sbi->ll_cache = NULL;
187 pcc_super_fini(&sbi->ll_pcc_super);
188 OBD_FREE(sbi, sizeof(*sbi));
193 static int client_common_fill_super(struct super_block *sb, char *md, char *dt,
194 struct vfsmount *mnt)
196 struct inode *root = NULL;
197 struct ll_sb_info *sbi = ll_s2sbi(sb);
198 struct obd_statfs *osfs = NULL;
199 struct ptlrpc_request *request = NULL;
200 struct obd_connect_data *data = NULL;
201 struct obd_uuid *uuid;
202 struct md_op_data *op_data;
203 struct lustre_md lmd;
205 int size, err, checksum;
208 sbi->ll_md_obd = class_name2obd(md);
209 if (!sbi->ll_md_obd) {
210 CERROR("MD %s: not setup or attached\n", md);
224 /* pass client page size via ocd_grant_blkbits, the server should report
225 * back its backend blocksize for grant calculation purpose */
226 data->ocd_grant_blkbits = PAGE_SHIFT;
228 /* indicate MDT features supported by this client */
229 data->ocd_connect_flags = OBD_CONNECT_IBITS | OBD_CONNECT_NODEVOH |
230 OBD_CONNECT_ATTRFID | OBD_CONNECT_GRANT |
231 OBD_CONNECT_VERSION | OBD_CONNECT_BRW_SIZE |
232 OBD_CONNECT_SRVLOCK | OBD_CONNECT_TRUNCLOCK|
233 OBD_CONNECT_MDS_CAPA | OBD_CONNECT_OSS_CAPA |
234 OBD_CONNECT_CANCELSET | OBD_CONNECT_FID |
235 OBD_CONNECT_AT | OBD_CONNECT_LOV_V3 |
236 OBD_CONNECT_VBR | OBD_CONNECT_FULL20 |
237 OBD_CONNECT_64BITHASH |
238 OBD_CONNECT_EINPROGRESS |
239 OBD_CONNECT_JOBSTATS | OBD_CONNECT_LVB_TYPE |
240 OBD_CONNECT_LAYOUTLOCK | OBD_CONNECT_PINGLESS|
241 OBD_CONNECT_MAX_EASIZE |
242 OBD_CONNECT_FLOCK_DEAD |
243 OBD_CONNECT_DISP_STRIPE | OBD_CONNECT_LFSCK |
244 OBD_CONNECT_OPEN_BY_FID |
245 OBD_CONNECT_DIR_STRIPE |
246 OBD_CONNECT_BULK_MBITS | OBD_CONNECT_CKSUM |
247 OBD_CONNECT_SUBTREE |
248 OBD_CONNECT_MULTIMODRPCS |
249 OBD_CONNECT_GRANT_PARAM |
250 OBD_CONNECT_SHORTIO | OBD_CONNECT_FLAGS2;
252 data->ocd_connect_flags2 = OBD_CONNECT2_DIR_MIGRATE |
253 OBD_CONNECT2_SUM_STATFS |
254 OBD_CONNECT2_OVERSTRIPING |
256 OBD_CONNECT2_LOCK_CONVERT |
257 OBD_CONNECT2_ARCHIVE_ID_ARRAY |
259 OBD_CONNECT2_ASYNC_DISCARD |
262 #ifdef HAVE_LRU_RESIZE_SUPPORT
263 if (sbi->ll_flags & LL_SBI_LRU_RESIZE)
264 data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
266 #ifdef CONFIG_LUSTRE_FS_POSIX_ACL
267 data->ocd_connect_flags |= OBD_CONNECT_ACL | OBD_CONNECT_UMASK |
268 OBD_CONNECT_LARGE_ACL;
271 data->ocd_cksum_types = obd_cksum_types_supported_client();
273 if (OBD_FAIL_CHECK(OBD_FAIL_MDC_LIGHTWEIGHT))
274 /* flag mdc connection as lightweight, only used for test
275 * purpose, use with care */
276 data->ocd_connect_flags |= OBD_CONNECT_LIGHTWEIGHT;
278 data->ocd_ibits_known = MDS_INODELOCK_FULL;
279 data->ocd_version = LUSTRE_VERSION_CODE;
281 if (sb->s_flags & SB_RDONLY)
282 data->ocd_connect_flags |= OBD_CONNECT_RDONLY;
283 if (sbi->ll_flags & LL_SBI_USER_XATTR)
284 data->ocd_connect_flags |= OBD_CONNECT_XATTR;
287 /* Setting this indicates we correctly support S_NOSEC (See kernel
288 * commit 9e1f1de02c2275d7172e18dc4e7c2065777611bf)
290 sb->s_flags |= SB_NOSEC;
293 if (sbi->ll_flags & LL_SBI_FLOCK)
294 sbi->ll_fop = &ll_file_operations_flock;
295 else if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
296 sbi->ll_fop = &ll_file_operations;
298 sbi->ll_fop = &ll_file_operations_noflock;
300 /* always ping even if server suppress_pings */
301 if (sbi->ll_flags & LL_SBI_ALWAYS_PING)
302 data->ocd_connect_flags &= ~OBD_CONNECT_PINGLESS;
304 obd_connect_set_secctx(data);
306 #if defined(CONFIG_SECURITY)
307 data->ocd_connect_flags2 |= OBD_CONNECT2_SELINUX_POLICY;
310 data->ocd_brw_size = MD_MAX_BRW_SIZE;
312 err = obd_connect(NULL, &sbi->ll_md_exp, sbi->ll_md_obd,
313 &sbi->ll_sb_uuid, data, sbi->ll_cache);
315 LCONSOLE_ERROR_MSG(0x14f, "An MDT (md %s) is performing "
316 "recovery, of which this client is not a "
317 "part. Please wait for recovery to complete,"
318 " abort, or time out.\n", md);
321 CERROR("cannot connect to %s: rc = %d\n", md, err);
325 sbi->ll_md_exp->exp_connect_data = *data;
327 err = obd_fid_init(sbi->ll_md_exp->exp_obd, sbi->ll_md_exp,
328 LUSTRE_SEQ_METADATA);
330 CERROR("%s: Can't init metadata layer FID infrastructure, "
331 "rc = %d\n", sbi->ll_md_exp->exp_obd->obd_name, err);
335 /* For mount, we only need fs info from MDT0, and also in DNE, it
336 * can make sure the client can be mounted as long as MDT0 is
338 err = obd_statfs(NULL, sbi->ll_md_exp, osfs,
339 ktime_get_seconds() - sbi->ll_statfs_max_age,
340 OBD_STATFS_FOR_MDT0);
342 GOTO(out_md_fid, err);
344 /* This needs to be after statfs to ensure connect has finished.
345 * Note that "data" does NOT contain the valid connect reply.
346 * If connecting to a 1.8 server there will be no LMV device, so
347 * we can access the MDC export directly and exp_connect_flags will
348 * be non-zero, but if accessing an upgraded 2.1 server it will
349 * have the correct flags filled in.
350 * XXX: fill in the LMV exp_connect_flags from MDC(s). */
351 valid = exp_connect_flags(sbi->ll_md_exp) & CLIENT_CONNECT_MDT_REQD;
352 if (exp_connect_flags(sbi->ll_md_exp) != 0 &&
353 valid != CLIENT_CONNECT_MDT_REQD) {
356 OBD_ALLOC_WAIT(buf, PAGE_SIZE);
357 obd_connect_flags2str(buf, PAGE_SIZE,
358 valid ^ CLIENT_CONNECT_MDT_REQD, 0, ",");
359 LCONSOLE_ERROR_MSG(0x170, "Server %s does not support "
360 "feature(s) needed for correct operation "
361 "of this client (%s). Please upgrade "
362 "server or downgrade client.\n",
363 sbi->ll_md_exp->exp_obd->obd_name, buf);
364 OBD_FREE(buf, PAGE_SIZE);
365 GOTO(out_md_fid, err = -EPROTO);
368 size = sizeof(*data);
369 err = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_CONN_DATA),
370 KEY_CONN_DATA, &size, data);
372 CERROR("%s: Get connect data failed: rc = %d\n",
373 sbi->ll_md_exp->exp_obd->obd_name, err);
374 GOTO(out_md_fid, err);
377 LASSERT(osfs->os_bsize);
378 sb->s_blocksize = osfs->os_bsize;
379 sb->s_blocksize_bits = log2(osfs->os_bsize);
380 sb->s_magic = LL_SUPER_MAGIC;
381 sb->s_maxbytes = MAX_LFS_FILESIZE;
382 sbi->ll_namelen = osfs->os_namelen;
383 sbi->ll_mnt.mnt = current->fs->root.mnt;
385 if ((sbi->ll_flags & LL_SBI_USER_XATTR) &&
386 !(data->ocd_connect_flags & OBD_CONNECT_XATTR)) {
387 LCONSOLE_INFO("Disabling user_xattr feature because "
388 "it is not supported on the server\n");
389 sbi->ll_flags &= ~LL_SBI_USER_XATTR;
392 if (data->ocd_connect_flags & OBD_CONNECT_ACL) {
394 sb->s_flags |= SB_POSIXACL;
396 sbi->ll_flags |= LL_SBI_ACL;
398 LCONSOLE_INFO("client wants to enable acl, but mdt not!\n");
400 sb->s_flags &= ~SB_POSIXACL;
402 sbi->ll_flags &= ~LL_SBI_ACL;
405 if (data->ocd_connect_flags & OBD_CONNECT_64BITHASH)
406 sbi->ll_flags |= LL_SBI_64BIT_HASH;
408 if (data->ocd_connect_flags & OBD_CONNECT_LAYOUTLOCK)
409 sbi->ll_flags |= LL_SBI_LAYOUT_LOCK;
411 if (obd_connect_has_secctx(data))
412 sbi->ll_flags |= LL_SBI_FILE_SECCTX;
414 if (data->ocd_ibits_known & MDS_INODELOCK_XATTR) {
415 if (!(data->ocd_connect_flags & OBD_CONNECT_MAX_EASIZE)) {
416 LCONSOLE_INFO("%s: disabling xattr cache due to "
417 "unknown maximum xattr size.\n", dt);
418 } else if (!sbi->ll_xattr_cache_set) {
419 /* If xattr_cache is already set (no matter 0 or 1)
420 * during processing llog, it won't be enabled here. */
421 sbi->ll_flags |= LL_SBI_XATTR_CACHE;
422 sbi->ll_xattr_cache_enabled = 1;
426 sbi->ll_dt_obd = class_name2obd(dt);
427 if (!sbi->ll_dt_obd) {
428 CERROR("DT %s: not setup or attached\n", dt);
429 GOTO(out_md_fid, err = -ENODEV);
432 /* pass client page size via ocd_grant_blkbits, the server should report
433 * back its backend blocksize for grant calculation purpose */
434 data->ocd_grant_blkbits = PAGE_SHIFT;
436 /* indicate OST features supported by this client */
437 data->ocd_connect_flags = OBD_CONNECT_GRANT | OBD_CONNECT_VERSION |
438 OBD_CONNECT_REQPORTAL | OBD_CONNECT_BRW_SIZE |
439 OBD_CONNECT_CANCELSET | OBD_CONNECT_FID |
440 OBD_CONNECT_SRVLOCK | OBD_CONNECT_TRUNCLOCK|
441 OBD_CONNECT_AT | OBD_CONNECT_OSS_CAPA |
442 OBD_CONNECT_VBR | OBD_CONNECT_FULL20 |
443 OBD_CONNECT_64BITHASH | OBD_CONNECT_MAXBYTES |
444 OBD_CONNECT_EINPROGRESS |
445 OBD_CONNECT_JOBSTATS | OBD_CONNECT_LVB_TYPE |
446 OBD_CONNECT_LAYOUTLOCK |
447 OBD_CONNECT_PINGLESS | OBD_CONNECT_LFSCK |
448 OBD_CONNECT_BULK_MBITS | OBD_CONNECT_SHORTIO |
449 OBD_CONNECT_FLAGS2 | OBD_CONNECT_GRANT_SHRINK;
451 /* The client currently advertises support for OBD_CONNECT_LOCKAHEAD_OLD so it
452 * can interoperate with an older version of lockahead which was released prior
453 * to landing in master. This support will be dropped when 2.13 development
454 * starts. At the point, we should not just drop the connect flag (below), we
455 * should also remove the support in the code.
457 * Removing it means a few things:
458 * 1. Remove this section here
459 * 2. Remove CEF_NONBLOCK in ll_file_lockahead()
460 * 3. Remove function exp_connect_lockahead_old
461 * 4. Remove LDLM_FL_LOCKAHEAD_OLD_RESERVED in lustre_dlm_flags.h
463 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 12, 50, 0)
464 data->ocd_connect_flags |= OBD_CONNECT_LOCKAHEAD_OLD;
467 data->ocd_connect_flags2 = OBD_CONNECT2_LOCKAHEAD;
469 if (!OBD_FAIL_CHECK(OBD_FAIL_OSC_CONNECT_GRANT_PARAM))
470 data->ocd_connect_flags |= OBD_CONNECT_GRANT_PARAM;
472 /* OBD_CONNECT_CKSUM should always be set, even if checksums are
473 * disabled by default, because it can still be enabled on the
474 * fly via /sys. As a consequence, we still need to come to an
475 * agreement on the supported algorithms at connect time
477 data->ocd_connect_flags |= OBD_CONNECT_CKSUM;
479 if (OBD_FAIL_CHECK(OBD_FAIL_OSC_CKSUM_ADLER_ONLY))
480 data->ocd_cksum_types = OBD_CKSUM_ADLER;
482 data->ocd_cksum_types = obd_cksum_types_supported_client();
484 #ifdef HAVE_LRU_RESIZE_SUPPORT
485 data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
487 /* always ping even if server suppress_pings */
488 if (sbi->ll_flags & LL_SBI_ALWAYS_PING)
489 data->ocd_connect_flags &= ~OBD_CONNECT_PINGLESS;
491 CDEBUG(D_RPCTRACE, "ocd_connect_flags: %#llx ocd_version: %d "
492 "ocd_grant: %d\n", data->ocd_connect_flags,
493 data->ocd_version, data->ocd_grant);
495 sbi->ll_dt_obd->obd_upcall.onu_owner = &sbi->ll_lco;
496 sbi->ll_dt_obd->obd_upcall.onu_upcall = cl_ocd_update;
498 data->ocd_brw_size = DT_MAX_BRW_SIZE;
500 err = obd_connect(NULL, &sbi->ll_dt_exp, sbi->ll_dt_obd,
501 &sbi->ll_sb_uuid, data, sbi->ll_cache);
503 LCONSOLE_ERROR_MSG(0x150, "An OST (dt %s) is performing "
504 "recovery, of which this client is not a "
505 "part. Please wait for recovery to "
506 "complete, abort, or time out.\n", dt);
509 CERROR("%s: Cannot connect to %s: rc = %d\n",
510 sbi->ll_dt_exp->exp_obd->obd_name, dt, err);
514 sbi->ll_dt_exp->exp_connect_data = *data;
516 /* Don't change value if it was specified in the config log */
517 if (sbi->ll_ra_info.ra_max_read_ahead_whole_pages == -1) {
518 sbi->ll_ra_info.ra_max_read_ahead_whole_pages =
519 max_t(unsigned long, SBI_DEFAULT_READAHEAD_WHOLE_MAX,
520 (data->ocd_brw_size >> PAGE_SHIFT));
521 if (sbi->ll_ra_info.ra_max_read_ahead_whole_pages >
522 sbi->ll_ra_info.ra_max_pages_per_file)
523 sbi->ll_ra_info.ra_max_read_ahead_whole_pages =
524 sbi->ll_ra_info.ra_max_pages_per_file;
527 err = obd_fid_init(sbi->ll_dt_exp->exp_obd, sbi->ll_dt_exp,
528 LUSTRE_SEQ_METADATA);
530 CERROR("%s: Can't init data layer FID infrastructure, "
531 "rc = %d\n", sbi->ll_dt_exp->exp_obd->obd_name, err);
535 mutex_lock(&sbi->ll_lco.lco_lock);
536 sbi->ll_lco.lco_flags = data->ocd_connect_flags;
537 sbi->ll_lco.lco_md_exp = sbi->ll_md_exp;
538 sbi->ll_lco.lco_dt_exp = sbi->ll_dt_exp;
539 mutex_unlock(&sbi->ll_lco.lco_lock);
541 fid_zero(&sbi->ll_root_fid);
542 err = md_get_root(sbi->ll_md_exp, get_mount_fileset(sb),
545 CERROR("cannot mds_connect: rc = %d\n", err);
546 GOTO(out_lock_cn_cb, err);
548 if (!fid_is_sane(&sbi->ll_root_fid)) {
549 CERROR("%s: Invalid root fid "DFID" during mount\n",
550 sbi->ll_md_exp->exp_obd->obd_name,
551 PFID(&sbi->ll_root_fid));
552 GOTO(out_lock_cn_cb, err = -EINVAL);
554 CDEBUG(D_SUPER, "rootfid "DFID"\n", PFID(&sbi->ll_root_fid));
556 sb->s_op = &lustre_super_operations;
557 sb->s_xattr = ll_xattr_handlers;
558 #if THREAD_SIZE >= 8192 /*b=17630*/
559 sb->s_export_op = &lustre_export_operations;
563 * XXX: move this to after cbd setup? */
564 valid = OBD_MD_FLGETATTR | OBD_MD_FLBLOCKS | OBD_MD_FLMODEASIZE;
565 if (sbi->ll_flags & LL_SBI_ACL)
566 valid |= OBD_MD_FLACL;
568 OBD_ALLOC_PTR(op_data);
570 GOTO(out_lock_cn_cb, err = -ENOMEM);
572 op_data->op_fid1 = sbi->ll_root_fid;
573 op_data->op_mode = 0;
574 op_data->op_valid = valid;
576 err = md_getattr(sbi->ll_md_exp, op_data, &request);
578 OBD_FREE_PTR(op_data);
580 CERROR("%s: md_getattr failed for root: rc = %d\n",
581 sbi->ll_md_exp->exp_obd->obd_name, err);
582 GOTO(out_lock_cn_cb, err);
585 err = md_get_lustre_md(sbi->ll_md_exp, request, sbi->ll_dt_exp,
586 sbi->ll_md_exp, &lmd);
588 CERROR("failed to understand root inode md: rc = %d\n", err);
589 ptlrpc_req_finished(request);
590 GOTO(out_lock_cn_cb, err);
593 LASSERT(fid_is_sane(&sbi->ll_root_fid));
594 root = ll_iget(sb, cl_fid_build_ino(&sbi->ll_root_fid,
595 sbi->ll_flags & LL_SBI_32BIT_API),
597 md_free_lustre_md(sbi->ll_md_exp, &lmd);
598 ptlrpc_req_finished(request);
601 #ifdef CONFIG_LUSTRE_FS_POSIX_ACL
603 posix_acl_release(lmd.posix_acl);
604 lmd.posix_acl = NULL;
607 err = IS_ERR(root) ? PTR_ERR(root) : -EBADF;
609 CERROR("lustre_lite: bad iget4 for root\n");
613 checksum = sbi->ll_flags & LL_SBI_CHECKSUM;
614 if (sbi->ll_checksum_set) {
615 err = obd_set_info_async(NULL, sbi->ll_dt_exp,
616 sizeof(KEY_CHECKSUM), KEY_CHECKSUM,
617 sizeof(checksum), &checksum, NULL);
619 CERROR("%s: Set checksum failed: rc = %d\n",
620 sbi->ll_dt_exp->exp_obd->obd_name, err);
626 sb->s_root = d_make_root(root);
627 if (sb->s_root == NULL) {
629 CERROR("%s: can't make root dentry: rc = %d\n",
630 sbi->ll_fsname, err);
634 sbi->ll_sdev_orig = sb->s_dev;
636 /* We set sb->s_dev equal on all lustre clients in order to support
637 * NFS export clustering. NFSD requires that the FSID be the same
639 /* s_dev is also used in lt_compare() to compare two fs, but that is
640 * only a node-local comparison. */
641 uuid = obd_get_uuid(sbi->ll_md_exp);
643 sb->s_dev = get_uuid2int(uuid->uuid, strlen(uuid->uuid));
650 if (sbi->ll_dt_obd) {
651 err = sysfs_create_link(&sbi->ll_kset.kobj,
652 &sbi->ll_dt_obd->obd_kset.kobj,
653 sbi->ll_dt_obd->obd_type->typ_name);
655 CERROR("%s: could not register %s in llite: rc = %d\n",
656 dt, sbi->ll_fsname, err);
661 if (sbi->ll_md_obd) {
662 err = sysfs_create_link(&sbi->ll_kset.kobj,
663 &sbi->ll_md_obd->obd_kset.kobj,
664 sbi->ll_md_obd->obd_type->typ_name);
666 CERROR("%s: could not register %s in llite: rc = %d\n",
667 md, sbi->ll_fsname, err);
677 obd_fid_fini(sbi->ll_dt_exp->exp_obd);
679 obd_disconnect(sbi->ll_dt_exp);
680 sbi->ll_dt_exp = NULL;
681 sbi->ll_dt_obd = NULL;
683 obd_fid_fini(sbi->ll_md_exp->exp_obd);
685 obd_disconnect(sbi->ll_md_exp);
686 sbi->ll_md_exp = NULL;
687 sbi->ll_md_obd = NULL;
696 int ll_get_max_mdsize(struct ll_sb_info *sbi, int *lmmsize)
700 size = sizeof(*lmmsize);
701 rc = obd_get_info(NULL, sbi->ll_dt_exp, sizeof(KEY_MAX_EASIZE),
702 KEY_MAX_EASIZE, &size, lmmsize);
704 CERROR("%s: cannot get max LOV EA size: rc = %d\n",
705 sbi->ll_dt_exp->exp_obd->obd_name, rc);
709 CDEBUG(D_INFO, "max LOV ea size: %d\n", *lmmsize);
712 rc = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_MAX_EASIZE),
713 KEY_MAX_EASIZE, &size, lmmsize);
715 CERROR("Get max mdsize error rc %d\n", rc);
717 CDEBUG(D_INFO, "max LMV ea size: %d\n", *lmmsize);
723 * Get the value of the default_easize parameter.
725 * \see client_obd::cl_default_mds_easize
727 * \param[in] sbi superblock info for this filesystem
728 * \param[out] lmmsize pointer to storage location for value
730 * \retval 0 on success
731 * \retval negative negated errno on failure
733 int ll_get_default_mdsize(struct ll_sb_info *sbi, int *lmmsize)
738 rc = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_DEFAULT_EASIZE),
739 KEY_DEFAULT_EASIZE, &size, lmmsize);
741 CERROR("Get default mdsize error rc %d\n", rc);
747 * Set the default_easize parameter to the given value.
749 * \see client_obd::cl_default_mds_easize
751 * \param[in] sbi superblock info for this filesystem
752 * \param[in] lmmsize the size to set
754 * \retval 0 on success
755 * \retval negative negated errno on failure
757 int ll_set_default_mdsize(struct ll_sb_info *sbi, int lmmsize)
761 if (lmmsize < sizeof(struct lov_mds_md) ||
762 lmmsize > OBD_MAX_DEFAULT_EA_SIZE)
765 rc = obd_set_info_async(NULL, sbi->ll_md_exp,
766 sizeof(KEY_DEFAULT_EASIZE), KEY_DEFAULT_EASIZE,
767 sizeof(int), &lmmsize, NULL);
772 static void client_common_put_super(struct super_block *sb)
774 struct ll_sb_info *sbi = ll_s2sbi(sb);
779 obd_fid_fini(sbi->ll_dt_exp->exp_obd);
780 obd_disconnect(sbi->ll_dt_exp);
781 sbi->ll_dt_exp = NULL;
783 ll_debugfs_unregister_super(sb);
785 obd_fid_fini(sbi->ll_md_exp->exp_obd);
786 obd_disconnect(sbi->ll_md_exp);
787 sbi->ll_md_exp = NULL;
792 void ll_kill_super(struct super_block *sb)
794 struct ll_sb_info *sbi;
798 if (!(sb->s_flags & SB_ACTIVE))
802 /* we need restore s_dev from changed for clustred NFS before put_super
803 * because new kernels have cached s_dev and change sb->s_dev in
804 * put_super not affected real removing devices */
806 sb->s_dev = sbi->ll_sdev_orig;
808 /* wait running statahead threads to quit */
809 while (atomic_read(&sbi->ll_sa_running) > 0) {
810 set_current_state(TASK_UNINTERRUPTIBLE);
811 schedule_timeout(cfs_time_seconds(1) >> 3);
818 static inline int ll_set_opt(const char *opt, char *data, int fl)
820 if (strncmp(opt, data, strlen(opt)) != 0)
826 /* non-client-specific mount options are parsed in lmd_parse */
827 static int ll_options(char *options, struct ll_sb_info *sbi)
830 char *s1 = options, *s2;
831 int *flags = &sbi->ll_flags;
837 CDEBUG(D_CONFIG, "Parsing opts %s\n", options);
840 CDEBUG(D_SUPER, "next opt=%s\n", s1);
841 tmp = ll_set_opt("nolock", s1, LL_SBI_NOLCK);
846 tmp = ll_set_opt("flock", s1, LL_SBI_FLOCK);
851 tmp = ll_set_opt("localflock", s1, LL_SBI_LOCALFLOCK);
856 tmp = ll_set_opt("noflock", s1, LL_SBI_FLOCK|LL_SBI_LOCALFLOCK);
861 tmp = ll_set_opt("user_xattr", s1, LL_SBI_USER_XATTR);
866 tmp = ll_set_opt("nouser_xattr", s1, LL_SBI_USER_XATTR);
871 tmp = ll_set_opt("context", s1, 1);
874 tmp = ll_set_opt("fscontext", s1, 1);
877 tmp = ll_set_opt("defcontext", s1, 1);
880 tmp = ll_set_opt("rootcontext", s1, 1);
883 tmp = ll_set_opt("user_fid2path", s1, LL_SBI_USER_FID2PATH);
888 tmp = ll_set_opt("nouser_fid2path", s1, LL_SBI_USER_FID2PATH);
894 tmp = ll_set_opt("checksum", s1, LL_SBI_CHECKSUM);
897 sbi->ll_checksum_set = 1;
900 tmp = ll_set_opt("nochecksum", s1, LL_SBI_CHECKSUM);
903 sbi->ll_checksum_set = 1;
906 tmp = ll_set_opt("lruresize", s1, LL_SBI_LRU_RESIZE);
911 tmp = ll_set_opt("nolruresize", s1, LL_SBI_LRU_RESIZE);
916 tmp = ll_set_opt("lazystatfs", s1, LL_SBI_LAZYSTATFS);
921 tmp = ll_set_opt("nolazystatfs", s1, LL_SBI_LAZYSTATFS);
926 tmp = ll_set_opt("32bitapi", s1, LL_SBI_32BIT_API);
931 tmp = ll_set_opt("verbose", s1, LL_SBI_VERBOSE);
936 tmp = ll_set_opt("noverbose", s1, LL_SBI_VERBOSE);
941 tmp = ll_set_opt("always_ping", s1, LL_SBI_ALWAYS_PING);
946 LCONSOLE_ERROR_MSG(0x152, "Unknown option '%s', won't mount.\n",
952 s2 = strchr(s1, ',');
960 void ll_lli_init(struct ll_inode_info *lli)
962 lli->lli_inode_magic = LLI_INODE_MAGIC;
964 spin_lock_init(&lli->lli_lock);
965 lli->lli_posix_acl = NULL;
966 /* Do not set lli_fid, it has been initialized already. */
967 fid_zero(&lli->lli_pfid);
968 lli->lli_mds_read_och = NULL;
969 lli->lli_mds_write_och = NULL;
970 lli->lli_mds_exec_och = NULL;
971 lli->lli_open_fd_read_count = 0;
972 lli->lli_open_fd_write_count = 0;
973 lli->lli_open_fd_exec_count = 0;
974 mutex_init(&lli->lli_och_mutex);
975 spin_lock_init(&lli->lli_agl_lock);
976 spin_lock_init(&lli->lli_layout_lock);
977 ll_layout_version_set(lli, CL_LAYOUT_GEN_NONE);
978 lli->lli_clob = NULL;
980 init_rwsem(&lli->lli_xattrs_list_rwsem);
981 mutex_init(&lli->lli_xattrs_enq_lock);
983 LASSERT(lli->lli_vfs_inode.i_mode != 0);
984 if (S_ISDIR(lli->lli_vfs_inode.i_mode)) {
985 lli->lli_opendir_key = NULL;
987 spin_lock_init(&lli->lli_sa_lock);
988 lli->lli_opendir_pid = 0;
989 lli->lli_sa_enabled = 0;
990 init_rwsem(&lli->lli_lsm_sem);
992 mutex_init(&lli->lli_size_mutex);
993 lli->lli_symlink_name = NULL;
994 init_rwsem(&lli->lli_trunc_sem);
995 range_lock_tree_init(&lli->lli_write_tree);
996 init_rwsem(&lli->lli_glimpse_sem);
997 lli->lli_glimpse_time = ktime_set(0, 0);
998 INIT_LIST_HEAD(&lli->lli_agl_list);
999 lli->lli_agl_index = 0;
1000 lli->lli_async_rc = 0;
1001 spin_lock_init(&lli->lli_heat_lock);
1002 obd_heat_clear(lli->lli_heat_instances, OBD_HEAT_COUNT);
1003 lli->lli_heat_flags = 0;
1004 mutex_init(&lli->lli_pcc_lock);
1005 lli->lli_pcc_state = PCC_STATE_FL_NONE;
1006 lli->lli_pcc_inode = NULL;
1007 mutex_init(&lli->lli_group_mutex);
1008 lli->lli_group_users = 0;
1009 lli->lli_group_gid = 0;
1011 mutex_init(&lli->lli_layout_mutex);
1012 memset(lli->lli_jobid, 0, sizeof(lli->lli_jobid));
1015 #define MAX_STRING_SIZE 128
1017 #ifndef HAVE_SUPER_SETUP_BDI_NAME
1019 #define LSI_BDI_INITIALIZED 0x00400000
1021 #ifndef HAVE_BDI_CAP_MAP_COPY
1022 # define BDI_CAP_MAP_COPY 0
1025 static int super_setup_bdi_name(struct super_block *sb, char *fmt, ...)
1027 struct lustre_sb_info *lsi = s2lsi(sb);
1028 char buf[MAX_STRING_SIZE];
1032 err = bdi_init(&lsi->lsi_bdi);
1036 lsi->lsi_flags |= LSI_BDI_INITIALIZED;
1037 lsi->lsi_bdi.capabilities = BDI_CAP_MAP_COPY;
1038 lsi->lsi_bdi.name = "lustre";
1039 va_start(args, fmt);
1040 vsnprintf(buf, MAX_STRING_SIZE, fmt, args);
1042 err = bdi_register(&lsi->lsi_bdi, NULL, "%s", buf);
1045 sb->s_bdi = &lsi->lsi_bdi;
1049 #endif /* !HAVE_SUPER_SETUP_BDI_NAME */
1051 int ll_fill_super(struct super_block *sb, struct vfsmount *mnt)
1053 struct lustre_profile *lprof = NULL;
1054 struct lustre_sb_info *lsi = s2lsi(sb);
1055 struct ll_sb_info *sbi = NULL;
1056 char *dt = NULL, *md = NULL;
1057 char *profilenm = get_profile_name(sb);
1058 struct config_llog_instance *cfg;
1059 /* %p for void* in printf needs 16+2 characters: 0xffffffffffffffff */
1060 const int instlen = 16 + 2;
1061 unsigned long cfg_instance = ll_get_cfg_instance(sb);
1062 char name[MAX_STRING_SIZE];
1071 /* for ASLR, to map between cfg_instance and hashed ptr */
1072 CDEBUG(D_VFSTRACE, "VFS Op: cfg_instance %s-%016lx (sb %p)\n",
1073 profilenm, cfg_instance, sb);
1075 try_module_get(THIS_MODULE);
1079 GOTO(out_free_cfg, err = -ENOMEM);
1081 /* client additional sb info */
1082 lsi->lsi_llsbi = sbi = ll_init_sbi();
1084 GOTO(out_free_cfg, err = PTR_ERR(sbi));
1086 err = ll_options(lsi->lsi_lmd->lmd_opts, sbi);
1088 GOTO(out_free_cfg, err);
1090 /* kernel >= 2.6.38 store dentry operations in sb->s_d_op. */
1091 sb->s_d_op = &ll_d_ops;
1094 generate_random_uuid(uuid.b);
1095 snprintf(sbi->ll_sb_uuid.uuid, UUID_SIZE, "%pU", uuid.b);
1097 CDEBUG(D_CONFIG, "llite sb uuid: %s\n", sbi->ll_sb_uuid.uuid);
1100 len = strlen(profilenm);
1101 ptr = strrchr(profilenm, '-');
1102 if (ptr && (strcmp(ptr, "-client") == 0))
1105 if (len > LUSTRE_MAXFSNAME) {
1106 if (unlikely(len >= MAX_STRING_SIZE))
1107 len = MAX_STRING_SIZE - 1;
1108 strncpy(name, profilenm, len);
1110 err = -ENAMETOOLONG;
1111 CERROR("%s: fsname longer than %u characters: rc = %d\n",
1112 name, LUSTRE_MAXFSNAME, err);
1113 GOTO(out_free_cfg, err);
1115 strncpy(sbi->ll_fsname, profilenm, len);
1116 sbi->ll_fsname[len] = '\0';
1119 snprintf(name, sizeof(name), "%.*s-%016lx", len,
1120 profilenm, cfg_instance);
1122 err = super_setup_bdi_name(sb, "%s", name);
1124 GOTO(out_free_cfg, err);
1126 /* Call ll_debugfs_register_super() before lustre_process_log()
1127 * so that "llite.*.*" params can be processed correctly.
1129 err = ll_debugfs_register_super(sb, name);
1131 CERROR("%s: could not register mountpoint in llite: rc = %d\n",
1132 sbi->ll_fsname, err);
1136 /* The cfg_instance is a value unique to this super, in case some
1137 * joker tries to mount the same fs at two mount points.
1139 cfg->cfg_instance = cfg_instance;
1140 cfg->cfg_uuid = lsi->lsi_llsbi->ll_sb_uuid;
1141 cfg->cfg_callback = class_config_llog_handler;
1142 cfg->cfg_sub_clds = CONFIG_SUB_CLIENT;
1143 /* set up client obds */
1144 err = lustre_process_log(sb, profilenm, cfg);
1146 GOTO(out_debugfs, err);
1148 /* Profile set with LCFG_MOUNTOPT so we can find our mdc and osc obds */
1149 lprof = class_get_profile(profilenm);
1150 if (lprof == NULL) {
1151 LCONSOLE_ERROR_MSG(0x156, "The client profile '%s' could not be"
1152 " read from the MGS. Does that filesystem "
1153 "exist?\n", profilenm);
1154 GOTO(out_debugfs, err = -EINVAL);
1156 CDEBUG(D_CONFIG, "Found profile %s: mdc=%s osc=%s\n", profilenm,
1157 lprof->lp_md, lprof->lp_dt);
1159 dt_len = strlen(lprof->lp_dt) + instlen + 2;
1160 OBD_ALLOC(dt, dt_len);
1162 GOTO(out_profile, err = -ENOMEM);
1163 snprintf(dt, dt_len - 1, "%s-%016lx", lprof->lp_dt, cfg_instance);
1165 md_len = strlen(lprof->lp_md) + instlen + 2;
1166 OBD_ALLOC(md, md_len);
1168 GOTO(out_free_dt, err = -ENOMEM);
1169 snprintf(md, md_len - 1, "%s-%016lx", lprof->lp_md, cfg_instance);
1171 /* connections, registrations, sb setup */
1172 err = client_common_fill_super(sb, md, dt, mnt);
1174 GOTO(out_free_md, err);
1176 sbi->ll_client_common_fill_super_succeeded = 1;
1180 OBD_FREE(md, md_len);
1183 OBD_FREE(dt, dt_len);
1186 class_put_profile(lprof);
1189 ll_debugfs_unregister_super(sb);
1196 else if (sbi->ll_flags & LL_SBI_VERBOSE)
1197 LCONSOLE_WARN("Mounted %s\n", profilenm);
1199 } /* ll_fill_super */
1201 void ll_put_super(struct super_block *sb)
1203 struct config_llog_instance cfg, params_cfg;
1204 struct obd_device *obd;
1205 struct lustre_sb_info *lsi = s2lsi(sb);
1206 struct ll_sb_info *sbi = ll_s2sbi(sb);
1207 char *profilenm = get_profile_name(sb);
1208 unsigned long cfg_instance = ll_get_cfg_instance(sb);
1210 int next, force = 1, rc = 0;
1214 GOTO(out_no_sbi, 0);
1216 /* Should replace instance_id with something better for ASLR */
1217 CDEBUG(D_VFSTRACE, "VFS Op: cfg_instance %s-%016lx (sb %p)\n",
1218 profilenm, cfg_instance, sb);
1220 cfg.cfg_instance = cfg_instance;
1221 lustre_end_log(sb, profilenm, &cfg);
1223 params_cfg.cfg_instance = cfg_instance;
1224 lustre_end_log(sb, PARAMS_FILENAME, ¶ms_cfg);
1226 if (sbi->ll_md_exp) {
1227 obd = class_exp2obd(sbi->ll_md_exp);
1229 force = obd->obd_force;
1232 /* Wait for unstable pages to be committed to stable storage */
1234 struct l_wait_info lwi = LWI_INTR(LWI_ON_SIGNAL_NOOP, NULL);
1235 rc = l_wait_event(sbi->ll_cache->ccc_unstable_waitq,
1236 atomic_long_read(&sbi->ll_cache->ccc_unstable_nr) == 0,
1240 ccc_count = atomic_long_read(&sbi->ll_cache->ccc_unstable_nr);
1241 if (force == 0 && rc != -EINTR)
1242 LASSERTF(ccc_count == 0, "count: %li\n", ccc_count);
1244 /* We need to set force before the lov_disconnect in
1245 lustre_common_put_super, since l_d cleans up osc's as well. */
1248 while ((obd = class_devices_in_group(&sbi->ll_sb_uuid,
1250 obd->obd_force = force;
1254 if (sbi->ll_client_common_fill_super_succeeded) {
1255 /* Only if client_common_fill_super succeeded */
1256 client_common_put_super(sb);
1260 while ((obd = class_devices_in_group(&sbi->ll_sb_uuid, &next)) !=NULL) {
1261 class_manual_cleanup(obd);
1264 if (sbi->ll_flags & LL_SBI_VERBOSE)
1265 LCONSOLE_WARN("Unmounted %s\n", profilenm ? profilenm : "");
1268 class_del_profile(profilenm);
1270 #ifndef HAVE_SUPER_SETUP_BDI_NAME
1271 if (lsi->lsi_flags & LSI_BDI_INITIALIZED) {
1272 bdi_destroy(&lsi->lsi_bdi);
1273 lsi->lsi_flags &= ~LSI_BDI_INITIALIZED;
1278 lsi->lsi_llsbi = NULL;
1280 lustre_common_put_super(sb);
1282 cl_env_cache_purge(~0);
1284 module_put(THIS_MODULE);
1287 } /* client_put_super */
1289 struct inode *ll_inode_from_resource_lock(struct ldlm_lock *lock)
1291 struct inode *inode = NULL;
1293 /* NOTE: we depend on atomic igrab() -bzzz */
1294 lock_res_and_lock(lock);
1295 if (lock->l_resource->lr_lvb_inode) {
1296 struct ll_inode_info * lli;
1297 lli = ll_i2info(lock->l_resource->lr_lvb_inode);
1298 if (lli->lli_inode_magic == LLI_INODE_MAGIC) {
1299 inode = igrab(lock->l_resource->lr_lvb_inode);
1301 inode = lock->l_resource->lr_lvb_inode;
1302 LDLM_DEBUG_LIMIT(inode->i_state & I_FREEING ? D_INFO :
1303 D_WARNING, lock, "lr_lvb_inode %p is "
1304 "bogus: magic %08x",
1305 lock->l_resource->lr_lvb_inode,
1306 lli->lli_inode_magic);
1310 unlock_res_and_lock(lock);
1314 void ll_dir_clear_lsm_md(struct inode *inode)
1316 struct ll_inode_info *lli = ll_i2info(inode);
1318 LASSERT(S_ISDIR(inode->i_mode));
1320 if (lli->lli_lsm_md) {
1321 lmv_free_memmd(lli->lli_lsm_md);
1322 lli->lli_lsm_md = NULL;
1325 if (lli->lli_default_lsm_md) {
1326 lmv_free_memmd(lli->lli_default_lsm_md);
1327 lli->lli_default_lsm_md = NULL;
1331 static struct inode *ll_iget_anon_dir(struct super_block *sb,
1332 const struct lu_fid *fid,
1333 struct lustre_md *md)
1335 struct ll_sb_info *sbi = ll_s2sbi(sb);
1336 struct mdt_body *body = md->body;
1337 struct inode *inode;
1341 ino = cl_fid_build_ino(fid, sbi->ll_flags & LL_SBI_32BIT_API);
1342 inode = iget_locked(sb, ino);
1343 if (inode == NULL) {
1344 CERROR("%s: failed get simple inode "DFID": rc = -ENOENT\n",
1345 sbi->ll_fsname, PFID(fid));
1346 RETURN(ERR_PTR(-ENOENT));
1349 if (inode->i_state & I_NEW) {
1350 struct ll_inode_info *lli = ll_i2info(inode);
1351 struct lmv_stripe_md *lsm = md->lmv;
1353 inode->i_mode = (inode->i_mode & ~S_IFMT) |
1354 (body->mbo_mode & S_IFMT);
1355 LASSERTF(S_ISDIR(inode->i_mode), "Not slave inode "DFID"\n",
1358 inode->i_mtime.tv_sec = 0;
1359 inode->i_atime.tv_sec = 0;
1360 inode->i_ctime.tv_sec = 0;
1363 #ifdef HAVE_BACKING_DEV_INFO
1364 /* initializing backing dev info. */
1365 inode->i_mapping->backing_dev_info =
1366 &s2lsi(inode->i_sb)->lsi_bdi;
1368 inode->i_op = &ll_dir_inode_operations;
1369 inode->i_fop = &ll_dir_operations;
1370 lli->lli_fid = *fid;
1373 LASSERT(lsm != NULL);
1374 /* master object FID */
1375 lli->lli_pfid = body->mbo_fid1;
1376 CDEBUG(D_INODE, "lli %p slave "DFID" master "DFID"\n",
1377 lli, PFID(fid), PFID(&lli->lli_pfid));
1378 unlock_new_inode(inode);
1384 static int ll_init_lsm_md(struct inode *inode, struct lustre_md *md)
1387 struct lmv_stripe_md *lsm = md->lmv;
1388 struct ll_inode_info *lli = ll_i2info(inode);
1391 LASSERT(lsm != NULL);
1393 CDEBUG(D_INODE, "%s: "DFID" set dir layout:\n",
1394 ll_i2sbi(inode)->ll_fsname, PFID(&lli->lli_fid));
1395 lsm_md_dump(D_INODE, lsm);
1397 if (!lmv_dir_striped(lsm))
1400 /* XXX sigh, this lsm_root initialization should be in
1401 * LMV layer, but it needs ll_iget right now, so we
1402 * put this here right now. */
1403 for (i = 0; i < lsm->lsm_md_stripe_count; i++) {
1404 fid = &lsm->lsm_md_oinfo[i].lmo_fid;
1405 LASSERT(lsm->lsm_md_oinfo[i].lmo_root == NULL);
1407 if (!fid_is_sane(fid))
1410 /* Unfortunately ll_iget will call ll_update_inode,
1411 * where the initialization of slave inode is slightly
1412 * different, so it reset lsm_md to NULL to avoid
1413 * initializing lsm for slave inode. */
1414 lsm->lsm_md_oinfo[i].lmo_root =
1415 ll_iget_anon_dir(inode->i_sb, fid, md);
1416 if (IS_ERR(lsm->lsm_md_oinfo[i].lmo_root)) {
1417 int rc = PTR_ERR(lsm->lsm_md_oinfo[i].lmo_root);
1419 lsm->lsm_md_oinfo[i].lmo_root = NULL;
1421 iput(lsm->lsm_md_oinfo[i].lmo_root);
1422 lsm->lsm_md_oinfo[i].lmo_root = NULL;
1428 lli->lli_lsm_md = lsm;
1433 static void ll_update_default_lsm_md(struct inode *inode, struct lustre_md *md)
1435 struct ll_inode_info *lli = ll_i2info(inode);
1437 if (!md->default_lmv) {
1438 /* clear default lsm */
1439 if (lli->lli_default_lsm_md) {
1440 down_write(&lli->lli_lsm_sem);
1441 if (lli->lli_default_lsm_md) {
1442 lmv_free_memmd(lli->lli_default_lsm_md);
1443 lli->lli_default_lsm_md = NULL;
1445 up_write(&lli->lli_lsm_sem);
1447 } else if (lli->lli_default_lsm_md) {
1448 /* update default lsm if it changes */
1449 down_read(&lli->lli_lsm_sem);
1450 if (lli->lli_default_lsm_md &&
1451 !lsm_md_eq(lli->lli_default_lsm_md, md->default_lmv)) {
1452 up_read(&lli->lli_lsm_sem);
1453 down_write(&lli->lli_lsm_sem);
1454 if (lli->lli_default_lsm_md)
1455 lmv_free_memmd(lli->lli_default_lsm_md);
1456 lli->lli_default_lsm_md = md->default_lmv;
1457 lsm_md_dump(D_INODE, md->default_lmv);
1458 md->default_lmv = NULL;
1459 up_write(&lli->lli_lsm_sem);
1461 up_read(&lli->lli_lsm_sem);
1464 /* init default lsm */
1465 down_write(&lli->lli_lsm_sem);
1466 lli->lli_default_lsm_md = md->default_lmv;
1467 lsm_md_dump(D_INODE, md->default_lmv);
1468 md->default_lmv = NULL;
1469 up_write(&lli->lli_lsm_sem);
1473 static int ll_update_lsm_md(struct inode *inode, struct lustre_md *md)
1475 struct ll_inode_info *lli = ll_i2info(inode);
1476 struct lmv_stripe_md *lsm = md->lmv;
1481 LASSERT(S_ISDIR(inode->i_mode));
1482 CDEBUG(D_INODE, "update lsm %p of "DFID"\n", lli->lli_lsm_md,
1483 PFID(ll_inode2fid(inode)));
1485 /* update default LMV */
1486 if (md->default_lmv)
1487 ll_update_default_lsm_md(inode, md);
1490 * no striped information from request, lustre_md from req does not
1491 * include stripeEA, see ll_md_setattr()
1497 * normally dir layout doesn't change, only take read lock to check
1498 * that to avoid blocking other MD operations.
1500 if (lli->lli_lsm_md)
1501 down_read(&lli->lli_lsm_sem);
1503 down_write(&lli->lli_lsm_sem);
1506 * if dir layout mismatch, check whether version is increased, which
1507 * means layout is changed, this happens in dir migration and lfsck.
1509 * foreign LMV should not change.
1511 if (lli->lli_lsm_md && !lsm_md_eq(lli->lli_lsm_md, lsm)) {
1512 if (lmv_dir_striped(lli->lli_lsm_md) &&
1513 lsm->lsm_md_layout_version <=
1514 lli->lli_lsm_md->lsm_md_layout_version) {
1515 CERROR("%s: "DFID" dir layout mismatch:\n",
1516 ll_i2sbi(inode)->ll_fsname,
1517 PFID(&lli->lli_fid));
1518 lsm_md_dump(D_ERROR, lli->lli_lsm_md);
1519 lsm_md_dump(D_ERROR, lsm);
1520 GOTO(unlock, rc = -EINVAL);
1523 /* layout changed, switch to write lock */
1524 up_read(&lli->lli_lsm_sem);
1525 down_write(&lli->lli_lsm_sem);
1526 ll_dir_clear_lsm_md(inode);
1529 /* set directory layout */
1530 if (!lli->lli_lsm_md) {
1531 struct cl_attr *attr;
1533 rc = ll_init_lsm_md(inode, md);
1534 up_write(&lli->lli_lsm_sem);
1538 /* set md->lmv to NULL, so the following free lustre_md
1539 * will not free this lsm */
1543 * md_merge_attr() may take long, since lsm is already set,
1544 * switch to read lock.
1546 down_read(&lli->lli_lsm_sem);
1548 if (!lmv_dir_striped(lli->lli_lsm_md))
1551 OBD_ALLOC_PTR(attr);
1553 GOTO(unlock, rc = -ENOMEM);
1555 /* validate the lsm */
1556 rc = md_merge_attr(ll_i2mdexp(inode), lsm, attr,
1557 ll_md_blocking_ast);
1563 if (md->body->mbo_valid & OBD_MD_FLNLINK)
1564 md->body->mbo_nlink = attr->cat_nlink;
1565 if (md->body->mbo_valid & OBD_MD_FLSIZE)
1566 md->body->mbo_size = attr->cat_size;
1567 if (md->body->mbo_valid & OBD_MD_FLATIME)
1568 md->body->mbo_atime = attr->cat_atime;
1569 if (md->body->mbo_valid & OBD_MD_FLCTIME)
1570 md->body->mbo_ctime = attr->cat_ctime;
1571 if (md->body->mbo_valid & OBD_MD_FLMTIME)
1572 md->body->mbo_mtime = attr->cat_mtime;
1577 up_read(&lli->lli_lsm_sem);
1582 void ll_clear_inode(struct inode *inode)
1584 struct ll_inode_info *lli = ll_i2info(inode);
1585 struct ll_sb_info *sbi = ll_i2sbi(inode);
1589 CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p)\n",
1590 PFID(ll_inode2fid(inode)), inode);
1592 if (S_ISDIR(inode->i_mode)) {
1593 /* these should have been cleared in ll_file_release */
1594 LASSERT(lli->lli_opendir_key == NULL);
1595 LASSERT(lli->lli_sai == NULL);
1596 LASSERT(lli->lli_opendir_pid == 0);
1598 pcc_inode_free(inode);
1601 md_null_inode(sbi->ll_md_exp, ll_inode2fid(inode));
1603 LASSERT(!lli->lli_open_fd_write_count);
1604 LASSERT(!lli->lli_open_fd_read_count);
1605 LASSERT(!lli->lli_open_fd_exec_count);
1607 if (lli->lli_mds_write_och)
1608 ll_md_real_close(inode, FMODE_WRITE);
1609 if (lli->lli_mds_exec_och)
1610 ll_md_real_close(inode, FMODE_EXEC);
1611 if (lli->lli_mds_read_och)
1612 ll_md_real_close(inode, FMODE_READ);
1614 if (S_ISLNK(inode->i_mode) && lli->lli_symlink_name) {
1615 OBD_FREE(lli->lli_symlink_name,
1616 strlen(lli->lli_symlink_name) + 1);
1617 lli->lli_symlink_name = NULL;
1620 ll_xattr_cache_destroy(inode);
1622 #ifdef CONFIG_LUSTRE_FS_POSIX_ACL
1623 forget_all_cached_acls(inode);
1624 if (lli->lli_posix_acl) {
1625 posix_acl_release(lli->lli_posix_acl);
1626 lli->lli_posix_acl = NULL;
1629 lli->lli_inode_magic = LLI_INODE_DEAD;
1631 if (S_ISDIR(inode->i_mode))
1632 ll_dir_clear_lsm_md(inode);
1633 else if (S_ISREG(inode->i_mode) && !is_bad_inode(inode))
1634 LASSERT(list_empty(&lli->lli_agl_list));
1637 * XXX This has to be done before lsm is freed below, because
1638 * cl_object still uses inode lsm.
1640 cl_inode_fini(inode);
1645 static int ll_md_setattr(struct dentry *dentry, struct md_op_data *op_data)
1647 struct lustre_md md;
1648 struct inode *inode = dentry->d_inode;
1649 struct ll_sb_info *sbi = ll_i2sbi(inode);
1650 struct ptlrpc_request *request = NULL;
1654 op_data = ll_prep_md_op_data(op_data, inode, NULL, NULL, 0, 0,
1655 LUSTRE_OPC_ANY, NULL);
1656 if (IS_ERR(op_data))
1657 RETURN(PTR_ERR(op_data));
1659 rc = md_setattr(sbi->ll_md_exp, op_data, NULL, 0, &request);
1661 ptlrpc_req_finished(request);
1662 if (rc == -ENOENT) {
1664 /* Unlinked special device node? Or just a race?
1665 * Pretend we done everything. */
1666 if (!S_ISREG(inode->i_mode) &&
1667 !S_ISDIR(inode->i_mode)) {
1668 ia_valid = op_data->op_attr.ia_valid;
1669 op_data->op_attr.ia_valid &= ~TIMES_SET_FLAGS;
1670 rc = simple_setattr(dentry, &op_data->op_attr);
1671 op_data->op_attr.ia_valid = ia_valid;
1673 } else if (rc != -EPERM && rc != -EACCES && rc != -ETXTBSY) {
1674 CERROR("md_setattr fails: rc = %d\n", rc);
1679 rc = md_get_lustre_md(sbi->ll_md_exp, request, sbi->ll_dt_exp,
1680 sbi->ll_md_exp, &md);
1682 ptlrpc_req_finished(request);
1686 ia_valid = op_data->op_attr.ia_valid;
1687 /* inode size will be in ll_setattr_ost, can't do it now since dirty
1688 * cache is not cleared yet. */
1689 op_data->op_attr.ia_valid &= ~(TIMES_SET_FLAGS | ATTR_SIZE);
1690 if (S_ISREG(inode->i_mode))
1692 rc = simple_setattr(dentry, &op_data->op_attr);
1693 if (S_ISREG(inode->i_mode))
1694 inode_unlock(inode);
1695 op_data->op_attr.ia_valid = ia_valid;
1697 rc = ll_update_inode(inode, &md);
1698 ptlrpc_req_finished(request);
1703 /* If this inode has objects allocated to it (lsm != NULL), then the OST
1704 * object(s) determine the file size and mtime. Otherwise, the MDS will
1705 * keep these values until such a time that objects are allocated for it.
1706 * We do the MDS operations first, as it is checking permissions for us.
1707 * We don't to the MDS RPC if there is nothing that we want to store there,
1708 * otherwise there is no harm in updating mtime/atime on the MDS if we are
1709 * going to do an RPC anyways.
1711 * If we are doing a truncate, we will send the mtime and ctime updates
1712 * to the OST with the punch RPC, otherwise we do an explicit setattr RPC.
1713 * I don't believe it is possible to get e.g. ATTR_MTIME_SET and ATTR_SIZE
1716 * In case of HSMimport, we only set attr on MDS.
1718 int ll_setattr_raw(struct dentry *dentry, struct iattr *attr,
1719 enum op_xvalid xvalid, bool hsm_import)
1721 struct inode *inode = dentry->d_inode;
1722 struct ll_inode_info *lli = ll_i2info(inode);
1723 struct md_op_data *op_data = NULL;
1728 CDEBUG(D_VFSTRACE, "%s: setattr inode "DFID"(%p) from %llu to %llu, "
1729 "valid %x, hsm_import %d\n",
1730 ll_i2sbi(inode)->ll_fsname, PFID(&lli->lli_fid),
1731 inode, i_size_read(inode), attr->ia_size, attr->ia_valid,
1734 if (attr->ia_valid & ATTR_SIZE) {
1735 /* Check new size against VFS/VM file size limit and rlimit */
1736 rc = inode_newsize_ok(inode, attr->ia_size);
1740 /* The maximum Lustre file size is variable, based on the
1741 * OST maximum object size and number of stripes. This
1742 * needs another check in addition to the VFS check above. */
1743 if (attr->ia_size > ll_file_maxbytes(inode)) {
1744 CDEBUG(D_INODE,"file "DFID" too large %llu > %llu\n",
1745 PFID(&lli->lli_fid), attr->ia_size,
1746 ll_file_maxbytes(inode));
1750 attr->ia_valid |= ATTR_MTIME | ATTR_CTIME;
1753 /* POSIX: check before ATTR_*TIME_SET set (from inode_change_ok) */
1754 if (attr->ia_valid & TIMES_SET_FLAGS) {
1755 if ((!uid_eq(current_fsuid(), inode->i_uid)) &&
1756 !cfs_capable(CFS_CAP_FOWNER))
1760 /* We mark all of the fields "set" so MDS/OST does not re-set them */
1761 if (!(xvalid & OP_XVALID_CTIME_SET) &&
1762 (attr->ia_valid & ATTR_CTIME)) {
1763 attr->ia_ctime = current_time(inode);
1764 xvalid |= OP_XVALID_CTIME_SET;
1766 if (!(attr->ia_valid & ATTR_ATIME_SET) &&
1767 (attr->ia_valid & ATTR_ATIME)) {
1768 attr->ia_atime = current_time(inode);
1769 attr->ia_valid |= ATTR_ATIME_SET;
1771 if (!(attr->ia_valid & ATTR_MTIME_SET) &&
1772 (attr->ia_valid & ATTR_MTIME)) {
1773 attr->ia_mtime = current_time(inode);
1774 attr->ia_valid |= ATTR_MTIME_SET;
1777 if (attr->ia_valid & (ATTR_MTIME | ATTR_CTIME))
1778 CDEBUG(D_INODE, "setting mtime %lld, ctime %lld, now = %lld\n",
1779 (s64)attr->ia_mtime.tv_sec, (s64)attr->ia_ctime.tv_sec,
1780 ktime_get_real_seconds());
1782 if (S_ISREG(inode->i_mode)) {
1783 if (attr->ia_valid & ATTR_SIZE)
1784 inode_dio_write_done(inode);
1785 inode_unlock(inode);
1788 /* We always do an MDS RPC, even if we're only changing the size;
1789 * only the MDS knows whether truncate() should fail with -ETXTBUSY */
1791 OBD_ALLOC_PTR(op_data);
1792 if (op_data == NULL)
1793 GOTO(out, rc = -ENOMEM);
1795 if (!hsm_import && attr->ia_valid & ATTR_SIZE) {
1796 /* If we are changing file size, file content is
1797 * modified, flag it.
1799 xvalid |= OP_XVALID_OWNEROVERRIDE;
1800 op_data->op_bias |= MDS_DATA_MODIFIED;
1801 ll_file_clear_flag(lli, LLIF_DATA_MODIFIED);
1804 if (attr->ia_valid & ATTR_FILE) {
1805 struct ll_file_data *fd = LUSTRE_FPRIVATE(attr->ia_file);
1807 if (fd->fd_lease_och)
1808 op_data->op_bias |= MDS_TRUNC_KEEP_LEASE;
1811 op_data->op_attr = *attr;
1812 op_data->op_xvalid = xvalid;
1814 rc = ll_md_setattr(dentry, op_data);
1818 if (!S_ISREG(inode->i_mode) || hsm_import)
1821 if (attr->ia_valid & (ATTR_SIZE | ATTR_ATIME | ATTR_ATIME_SET |
1822 ATTR_MTIME | ATTR_MTIME_SET | ATTR_CTIME) ||
1823 xvalid & OP_XVALID_CTIME_SET) {
1824 bool cached = false;
1826 rc = pcc_inode_setattr(inode, attr, &cached);
1829 CERROR("%s: PCC inode "DFID" setattr failed: "
1831 ll_i2sbi(inode)->ll_fsname,
1832 PFID(&lli->lli_fid), rc);
1836 /* For truncate and utimes sending attributes to OSTs,
1837 * setting mtime/atime to the past will be performed
1838 * under PW [0:EOF] extent lock (new_size:EOF for
1839 * truncate). It may seem excessive to send mtime/atime
1840 * updates to OSTs when not setting times to past, but
1841 * it is necessary due to possible time
1842 * de-synchronization between MDT inode and OST objects
1844 rc = cl_setattr_ost(lli->lli_clob, attr, xvalid, 0);
1848 /* If the file was restored, it needs to set dirty flag.
1850 * We've already sent MDS_DATA_MODIFIED flag in
1851 * ll_md_setattr() for truncate. However, the MDT refuses to
1852 * set the HS_DIRTY flag on released files, so we have to set
1853 * it again if the file has been restored. Please check how
1854 * LLIF_DATA_MODIFIED is set in vvp_io_setattr_fini().
1856 * Please notice that if the file is not released, the previous
1857 * MDS_DATA_MODIFIED has taken effect and usually
1858 * LLIF_DATA_MODIFIED is not set(see vvp_io_setattr_fini()).
1859 * This way we can save an RPC for common open + trunc
1861 if (ll_file_test_and_clear_flag(lli, LLIF_DATA_MODIFIED)) {
1862 struct hsm_state_set hss = {
1863 .hss_valid = HSS_SETMASK,
1864 .hss_setmask = HS_DIRTY,
1868 rc2 = ll_hsm_state_set(inode, &hss);
1869 /* truncate and write can happen at the same time, so that
1870 * the file can be set modified even though the file is not
1871 * restored from released state, and ll_hsm_state_set() is
1872 * not applicable for the file, and rc2 < 0 is normal in this
1875 CDEBUG(D_INFO, DFID "HSM set dirty failed: rc2 = %d\n",
1876 PFID(ll_inode2fid(inode)), rc2);
1881 if (op_data != NULL)
1882 ll_finish_md_op_data(op_data);
1884 if (S_ISREG(inode->i_mode)) {
1886 if ((attr->ia_valid & ATTR_SIZE) && !hsm_import)
1887 inode_dio_wait(inode);
1888 /* Once we've got the i_mutex, it's safe to set the S_NOSEC
1889 * flag. ll_update_inode (called from ll_md_setattr), clears
1890 * inode flags, so there is a gap where S_NOSEC is not set.
1891 * This can cause a writer to take the i_mutex unnecessarily,
1892 * but this is safe to do and should be rare. */
1893 inode_has_no_xattr(inode);
1896 ll_stats_ops_tally(ll_i2sbi(inode), (attr->ia_valid & ATTR_SIZE) ?
1897 LPROC_LL_TRUNC : LPROC_LL_SETATTR, 1);
1902 int ll_setattr(struct dentry *de, struct iattr *attr)
1904 int mode = de->d_inode->i_mode;
1905 enum op_xvalid xvalid = 0;
1907 if ((attr->ia_valid & (ATTR_CTIME|ATTR_SIZE|ATTR_MODE)) ==
1908 (ATTR_CTIME|ATTR_SIZE|ATTR_MODE))
1909 xvalid |= OP_XVALID_OWNEROVERRIDE;
1911 if (((attr->ia_valid & (ATTR_MODE|ATTR_FORCE|ATTR_SIZE)) ==
1912 (ATTR_SIZE|ATTR_MODE)) &&
1913 (((mode & S_ISUID) && !(attr->ia_mode & S_ISUID)) ||
1914 (((mode & (S_ISGID|S_IXGRP)) == (S_ISGID|S_IXGRP)) &&
1915 !(attr->ia_mode & S_ISGID))))
1916 attr->ia_valid |= ATTR_FORCE;
1918 if ((attr->ia_valid & ATTR_MODE) &&
1920 !(attr->ia_mode & S_ISUID) &&
1921 !(attr->ia_valid & ATTR_KILL_SUID))
1922 attr->ia_valid |= ATTR_KILL_SUID;
1924 if ((attr->ia_valid & ATTR_MODE) &&
1925 ((mode & (S_ISGID|S_IXGRP)) == (S_ISGID|S_IXGRP)) &&
1926 !(attr->ia_mode & S_ISGID) &&
1927 !(attr->ia_valid & ATTR_KILL_SGID))
1928 attr->ia_valid |= ATTR_KILL_SGID;
1930 return ll_setattr_raw(de, attr, xvalid, false);
1933 int ll_statfs_internal(struct ll_sb_info *sbi, struct obd_statfs *osfs,
1936 struct obd_statfs obd_osfs = { 0 };
1941 max_age = ktime_get_seconds() - sbi->ll_statfs_max_age;
1943 rc = obd_statfs(NULL, sbi->ll_md_exp, osfs, max_age, flags);
1947 osfs->os_type = LL_SUPER_MAGIC;
1949 CDEBUG(D_SUPER, "MDC blocks %llu/%llu objects %llu/%llu\n",
1950 osfs->os_bavail, osfs->os_blocks, osfs->os_ffree, osfs->os_files);
1952 if (osfs->os_state & OS_STATE_SUM)
1955 if (sbi->ll_flags & LL_SBI_LAZYSTATFS)
1956 flags |= OBD_STATFS_NODELAY;
1958 rc = obd_statfs(NULL, sbi->ll_dt_exp, &obd_osfs, max_age, flags);
1959 if (rc) /* Possibly a filesystem with no OSTs. Report MDT totals. */
1962 CDEBUG(D_SUPER, "OSC blocks %llu/%llu objects %llu/%llu\n",
1963 obd_osfs.os_bavail, obd_osfs.os_blocks, obd_osfs.os_ffree,
1966 osfs->os_bsize = obd_osfs.os_bsize;
1967 osfs->os_blocks = obd_osfs.os_blocks;
1968 osfs->os_bfree = obd_osfs.os_bfree;
1969 osfs->os_bavail = obd_osfs.os_bavail;
1971 /* If we have _some_ OSTs, but don't have as many free objects on the
1972 * OSTs as inodes on the MDTs, reduce the reported number of inodes
1973 * to compensate, so that the "inodes in use" number is correct.
1974 * This should be kept in sync with lod_statfs() behaviour.
1976 if (obd_osfs.os_files && obd_osfs.os_ffree < osfs->os_ffree) {
1977 osfs->os_files = (osfs->os_files - osfs->os_ffree) +
1979 osfs->os_ffree = obd_osfs.os_ffree;
1985 int ll_statfs(struct dentry *de, struct kstatfs *sfs)
1987 struct super_block *sb = de->d_sb;
1988 struct obd_statfs osfs;
1989 __u64 fsid = huge_encode_dev(sb->s_dev);
1992 CDEBUG(D_VFSTRACE, "VFS Op: at %llu jiffies\n", get_jiffies_64());
1993 ll_stats_ops_tally(ll_s2sbi(sb), LPROC_LL_STATFS, 1);
1995 /* Some amount of caching on the client is allowed */
1996 rc = ll_statfs_internal(ll_s2sbi(sb), &osfs, OBD_STATFS_SUM);
2000 statfs_unpack(sfs, &osfs);
2002 /* We need to downshift for all 32-bit kernels, because we can't
2003 * tell if the kernel is being called via sys_statfs64() or not.
2004 * Stop before overflowing f_bsize - in which case it is better
2005 * to just risk EOVERFLOW if caller is using old sys_statfs(). */
2006 if (sizeof(long) < 8) {
2007 while (osfs.os_blocks > ~0UL && sfs->f_bsize < 0x40000000) {
2010 osfs.os_blocks >>= 1;
2011 osfs.os_bfree >>= 1;
2012 osfs.os_bavail >>= 1;
2016 sfs->f_blocks = osfs.os_blocks;
2017 sfs->f_bfree = osfs.os_bfree;
2018 sfs->f_bavail = osfs.os_bavail;
2019 sfs->f_fsid.val[0] = (__u32)fsid;
2020 sfs->f_fsid.val[1] = (__u32)(fsid >> 32);
2024 void ll_inode_size_lock(struct inode *inode)
2026 struct ll_inode_info *lli;
2028 LASSERT(!S_ISDIR(inode->i_mode));
2030 lli = ll_i2info(inode);
2031 mutex_lock(&lli->lli_size_mutex);
2034 void ll_inode_size_unlock(struct inode *inode)
2036 struct ll_inode_info *lli;
2038 lli = ll_i2info(inode);
2039 mutex_unlock(&lli->lli_size_mutex);
2042 void ll_update_inode_flags(struct inode *inode, int ext_flags)
2044 inode->i_flags = ll_ext_to_inode_flags(ext_flags);
2045 if (ext_flags & LUSTRE_PROJINHERIT_FL)
2046 ll_file_set_flag(ll_i2info(inode), LLIF_PROJECT_INHERIT);
2048 ll_file_clear_flag(ll_i2info(inode), LLIF_PROJECT_INHERIT);
2051 int ll_update_inode(struct inode *inode, struct lustre_md *md)
2053 struct ll_inode_info *lli = ll_i2info(inode);
2054 struct mdt_body *body = md->body;
2055 struct ll_sb_info *sbi = ll_i2sbi(inode);
2058 if (body->mbo_valid & OBD_MD_FLEASIZE) {
2059 rc = cl_file_inode_init(inode, md);
2064 if (S_ISDIR(inode->i_mode)) {
2065 rc = ll_update_lsm_md(inode, md);
2070 #ifdef CONFIG_LUSTRE_FS_POSIX_ACL
2071 if (body->mbo_valid & OBD_MD_FLACL) {
2072 spin_lock(&lli->lli_lock);
2073 if (lli->lli_posix_acl)
2074 posix_acl_release(lli->lli_posix_acl);
2075 lli->lli_posix_acl = md->posix_acl;
2076 spin_unlock(&lli->lli_lock);
2079 inode->i_ino = cl_fid_build_ino(&body->mbo_fid1,
2080 sbi->ll_flags & LL_SBI_32BIT_API);
2081 inode->i_generation = cl_fid_build_gen(&body->mbo_fid1);
2083 if (body->mbo_valid & OBD_MD_FLATIME) {
2084 if (body->mbo_atime > inode->i_atime.tv_sec)
2085 inode->i_atime.tv_sec = body->mbo_atime;
2086 lli->lli_atime = body->mbo_atime;
2089 if (body->mbo_valid & OBD_MD_FLMTIME) {
2090 if (body->mbo_mtime > inode->i_mtime.tv_sec) {
2092 "setting ino %lu mtime from %lld to %llu\n",
2093 inode->i_ino, (s64)inode->i_mtime.tv_sec,
2095 inode->i_mtime.tv_sec = body->mbo_mtime;
2097 lli->lli_mtime = body->mbo_mtime;
2100 if (body->mbo_valid & OBD_MD_FLCTIME) {
2101 if (body->mbo_ctime > inode->i_ctime.tv_sec)
2102 inode->i_ctime.tv_sec = body->mbo_ctime;
2103 lli->lli_ctime = body->mbo_ctime;
2106 /* Clear i_flags to remove S_NOSEC before permissions are updated */
2107 if (body->mbo_valid & OBD_MD_FLFLAGS)
2108 ll_update_inode_flags(inode, body->mbo_flags);
2109 if (body->mbo_valid & OBD_MD_FLMODE)
2110 inode->i_mode = (inode->i_mode & S_IFMT) |
2111 (body->mbo_mode & ~S_IFMT);
2113 if (body->mbo_valid & OBD_MD_FLTYPE)
2114 inode->i_mode = (inode->i_mode & ~S_IFMT) |
2115 (body->mbo_mode & S_IFMT);
2117 LASSERT(inode->i_mode != 0);
2118 if (S_ISREG(inode->i_mode))
2119 inode->i_blkbits = min(PTLRPC_MAX_BRW_BITS + 1,
2120 LL_MAX_BLKSIZE_BITS);
2122 inode->i_blkbits = inode->i_sb->s_blocksize_bits;
2124 if (body->mbo_valid & OBD_MD_FLUID)
2125 inode->i_uid = make_kuid(&init_user_ns, body->mbo_uid);
2126 if (body->mbo_valid & OBD_MD_FLGID)
2127 inode->i_gid = make_kgid(&init_user_ns, body->mbo_gid);
2128 if (body->mbo_valid & OBD_MD_FLPROJID)
2129 lli->lli_projid = body->mbo_projid;
2130 if (body->mbo_valid & OBD_MD_FLNLINK)
2131 set_nlink(inode, body->mbo_nlink);
2132 if (body->mbo_valid & OBD_MD_FLRDEV)
2133 inode->i_rdev = old_decode_dev(body->mbo_rdev);
2135 if (body->mbo_valid & OBD_MD_FLID) {
2136 /* FID shouldn't be changed! */
2137 if (fid_is_sane(&lli->lli_fid)) {
2138 LASSERTF(lu_fid_eq(&lli->lli_fid, &body->mbo_fid1),
2139 "Trying to change FID "DFID
2140 " to the "DFID", inode "DFID"(%p)\n",
2141 PFID(&lli->lli_fid), PFID(&body->mbo_fid1),
2142 PFID(ll_inode2fid(inode)), inode);
2144 lli->lli_fid = body->mbo_fid1;
2148 LASSERT(fid_seq(&lli->lli_fid) != 0);
2150 if (body->mbo_valid & OBD_MD_FLSIZE) {
2151 i_size_write(inode, body->mbo_size);
2153 CDEBUG(D_VFSTRACE, "inode="DFID", updating i_size %llu\n",
2154 PFID(ll_inode2fid(inode)),
2155 (unsigned long long)body->mbo_size);
2157 if (body->mbo_valid & OBD_MD_FLBLOCKS)
2158 inode->i_blocks = body->mbo_blocks;
2161 if (body->mbo_valid & OBD_MD_TSTATE) {
2162 /* Set LLIF_FILE_RESTORING if restore ongoing and
2163 * clear it when done to ensure to start again
2164 * glimpsing updated attrs
2166 if (body->mbo_t_state & MS_RESTORE)
2167 ll_file_set_flag(lli, LLIF_FILE_RESTORING);
2169 ll_file_clear_flag(lli, LLIF_FILE_RESTORING);
2175 int ll_read_inode2(struct inode *inode, void *opaque)
2177 struct lustre_md *md = opaque;
2178 struct ll_inode_info *lli = ll_i2info(inode);
2182 CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p)\n",
2183 PFID(&lli->lli_fid), inode);
2185 /* Core attributes from the MDS first. This is a new inode, and
2186 * the VFS doesn't zero times in the core inode so we have to do
2187 * it ourselves. They will be overwritten by either MDS or OST
2188 * attributes - we just need to make sure they aren't newer.
2190 inode->i_mtime.tv_sec = 0;
2191 inode->i_atime.tv_sec = 0;
2192 inode->i_ctime.tv_sec = 0;
2194 rc = ll_update_inode(inode, md);
2198 /* OIDEBUG(inode); */
2200 #ifdef HAVE_BACKING_DEV_INFO
2201 /* initializing backing dev info. */
2202 inode->i_mapping->backing_dev_info = &s2lsi(inode->i_sb)->lsi_bdi;
2204 if (S_ISREG(inode->i_mode)) {
2205 struct ll_sb_info *sbi = ll_i2sbi(inode);
2206 inode->i_op = &ll_file_inode_operations;
2207 inode->i_fop = sbi->ll_fop;
2208 inode->i_mapping->a_ops = (struct address_space_operations *)&ll_aops;
2210 } else if (S_ISDIR(inode->i_mode)) {
2211 inode->i_op = &ll_dir_inode_operations;
2212 inode->i_fop = &ll_dir_operations;
2214 } else if (S_ISLNK(inode->i_mode)) {
2215 inode->i_op = &ll_fast_symlink_inode_operations;
2218 inode->i_op = &ll_special_inode_operations;
2220 init_special_inode(inode, inode->i_mode,
2229 void ll_delete_inode(struct inode *inode)
2231 struct ll_inode_info *lli = ll_i2info(inode);
2232 struct address_space *mapping = &inode->i_data;
2233 unsigned long nrpages;
2236 if (S_ISREG(inode->i_mode) && lli->lli_clob != NULL) {
2237 /* It is last chance to write out dirty pages,
2238 * otherwise we may lose data while umount.
2240 * If i_nlink is 0 then just discard data. This is safe because
2241 * local inode gets i_nlink 0 from server only for the last
2242 * unlink, so that file is not opened somewhere else
2244 cl_sync_file_range(inode, 0, OBD_OBJECT_EOF, inode->i_nlink ?
2245 CL_FSYNC_LOCAL : CL_FSYNC_DISCARD, 1);
2247 truncate_inode_pages_final(mapping);
2249 /* Workaround for LU-118: Note nrpages may not be totally updated when
2250 * truncate_inode_pages() returns, as there can be a page in the process
2251 * of deletion (inside __delete_from_page_cache()) in the specified
2252 * range. Thus mapping->nrpages can be non-zero when this function
2253 * returns even after truncation of the whole mapping. Only do this if
2254 * npages isn't already zero.
2256 nrpages = mapping->nrpages;
2258 xa_lock_irq(&mapping->i_pages);
2259 nrpages = mapping->nrpages;
2260 xa_unlock_irq(&mapping->i_pages);
2261 } /* Workaround end */
2263 LASSERTF(nrpages == 0, "%s: inode="DFID"(%p) nrpages=%lu, "
2264 "see https://jira.whamcloud.com/browse/LU-118\n",
2265 ll_i2sbi(inode)->ll_fsname,
2266 PFID(ll_inode2fid(inode)), inode, nrpages);
2268 ll_clear_inode(inode);
2274 int ll_iocontrol(struct inode *inode, struct file *file,
2275 unsigned int cmd, unsigned long arg)
2277 struct ll_sb_info *sbi = ll_i2sbi(inode);
2278 struct ptlrpc_request *req = NULL;
2283 case FS_IOC_GETFLAGS: {
2284 struct mdt_body *body;
2285 struct md_op_data *op_data;
2287 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL,
2288 0, 0, LUSTRE_OPC_ANY,
2290 if (IS_ERR(op_data))
2291 RETURN(PTR_ERR(op_data));
2293 op_data->op_valid = OBD_MD_FLFLAGS;
2294 rc = md_getattr(sbi->ll_md_exp, op_data, &req);
2295 ll_finish_md_op_data(op_data);
2297 CERROR("%s: failure inode "DFID": rc = %d\n",
2298 sbi->ll_md_exp->exp_obd->obd_name,
2299 PFID(ll_inode2fid(inode)), rc);
2303 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
2305 flags = body->mbo_flags;
2307 ptlrpc_req_finished(req);
2309 RETURN(put_user(flags, (int __user *)arg));
2311 case FS_IOC_SETFLAGS: {
2313 struct md_op_data *op_data;
2314 struct cl_object *obj;
2315 struct fsxattr fa = { 0 };
2317 if (get_user(flags, (int __user *)arg))
2320 fa.fsx_projid = ll_i2info(inode)->lli_projid;
2321 if (flags & LUSTRE_PROJINHERIT_FL)
2322 fa.fsx_xflags = FS_XFLAG_PROJINHERIT;
2324 rc = ll_ioctl_check_project(inode, &fa);
2328 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
2329 LUSTRE_OPC_ANY, NULL);
2330 if (IS_ERR(op_data))
2331 RETURN(PTR_ERR(op_data));
2333 op_data->op_attr_flags = flags;
2334 op_data->op_xvalid |= OP_XVALID_FLAGS;
2335 rc = md_setattr(sbi->ll_md_exp, op_data, NULL, 0, &req);
2336 ll_finish_md_op_data(op_data);
2337 ptlrpc_req_finished(req);
2341 ll_update_inode_flags(inode, flags);
2343 obj = ll_i2info(inode)->lli_clob;
2347 OBD_ALLOC_PTR(attr);
2351 rc = cl_setattr_ost(obj, attr, OP_XVALID_FLAGS, flags);
2363 int ll_flush_ctx(struct inode *inode)
2365 struct ll_sb_info *sbi = ll_i2sbi(inode);
2367 CDEBUG(D_SEC, "flush context for user %d\n",
2368 from_kuid(&init_user_ns, current_uid()));
2370 obd_set_info_async(NULL, sbi->ll_md_exp,
2371 sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
2373 obd_set_info_async(NULL, sbi->ll_dt_exp,
2374 sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
2379 /* umount -f client means force down, don't save state */
2380 void ll_umount_begin(struct super_block *sb)
2382 struct ll_sb_info *sbi = ll_s2sbi(sb);
2383 struct obd_device *obd;
2384 struct obd_ioctl_data *ioc_data;
2385 struct l_wait_info lwi;
2386 wait_queue_head_t waitq;
2389 CDEBUG(D_VFSTRACE, "VFS Op: superblock %p count %d active %d\n", sb,
2390 sb->s_count, atomic_read(&sb->s_active));
2392 obd = class_exp2obd(sbi->ll_md_exp);
2394 CERROR("Invalid MDC connection handle %#llx\n",
2395 sbi->ll_md_exp->exp_handle.h_cookie);
2401 obd = class_exp2obd(sbi->ll_dt_exp);
2403 CERROR("Invalid LOV connection handle %#llx\n",
2404 sbi->ll_dt_exp->exp_handle.h_cookie);
2410 OBD_ALLOC_PTR(ioc_data);
2412 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_md_exp,
2413 sizeof *ioc_data, ioc_data, NULL);
2415 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_dt_exp,
2416 sizeof *ioc_data, ioc_data, NULL);
2418 OBD_FREE_PTR(ioc_data);
2421 /* Really, we'd like to wait until there are no requests outstanding,
2422 * and then continue. For now, we just periodically checking for vfs
2423 * to decrement mnt_cnt and hope to finish it within 10sec.
2425 init_waitqueue_head(&waitq);
2426 lwi = LWI_TIMEOUT_INTERVAL(cfs_time_seconds(10),
2427 cfs_time_seconds(1), NULL, NULL);
2428 l_wait_event(waitq, may_umount(sbi->ll_mnt.mnt), &lwi);
2433 int ll_remount_fs(struct super_block *sb, int *flags, char *data)
2435 struct ll_sb_info *sbi = ll_s2sbi(sb);
2436 char *profilenm = get_profile_name(sb);
2440 if ((*flags & MS_RDONLY) != (sb->s_flags & SB_RDONLY)) {
2441 read_only = *flags & MS_RDONLY;
2442 err = obd_set_info_async(NULL, sbi->ll_md_exp,
2443 sizeof(KEY_READ_ONLY),
2444 KEY_READ_ONLY, sizeof(read_only),
2447 LCONSOLE_WARN("Failed to remount %s %s (%d)\n",
2448 profilenm, read_only ?
2449 "read-only" : "read-write", err);
2454 sb->s_flags |= SB_RDONLY;
2456 sb->s_flags &= ~SB_RDONLY;
2458 if (sbi->ll_flags & LL_SBI_VERBOSE)
2459 LCONSOLE_WARN("Remounted %s %s\n", profilenm,
2460 read_only ? "read-only" : "read-write");
2466 * Cleanup the open handle that is cached on MDT-side.
2468 * For open case, the client side open handling thread may hit error
2469 * after the MDT grant the open. Under such case, the client should
2470 * send close RPC to the MDT as cleanup; otherwise, the open handle
2471 * on the MDT will be leaked there until the client umount or evicted.
2473 * In further, if someone unlinked the file, because the open handle
2474 * holds the reference on such file/object, then it will block the
2475 * subsequent threads that want to locate such object via FID.
2477 * \param[in] sb super block for this file-system
2478 * \param[in] open_req pointer to the original open request
2480 void ll_open_cleanup(struct super_block *sb, struct ptlrpc_request *open_req)
2482 struct mdt_body *body;
2483 struct md_op_data *op_data;
2484 struct ptlrpc_request *close_req = NULL;
2485 struct obd_export *exp = ll_s2sbi(sb)->ll_md_exp;
2488 body = req_capsule_server_get(&open_req->rq_pill, &RMF_MDT_BODY);
2489 OBD_ALLOC_PTR(op_data);
2490 if (op_data == NULL) {
2491 CWARN("%s: cannot allocate op_data to release open handle for "
2492 DFID"\n", ll_s2sbi(sb)->ll_fsname, PFID(&body->mbo_fid1));
2497 op_data->op_fid1 = body->mbo_fid1;
2498 op_data->op_open_handle = body->mbo_open_handle;
2499 op_data->op_mod_time = ktime_get_real_seconds();
2500 md_close(exp, op_data, NULL, &close_req);
2501 ptlrpc_req_finished(close_req);
2502 ll_finish_md_op_data(op_data);
2507 int ll_prep_inode(struct inode **inode, struct ptlrpc_request *req,
2508 struct super_block *sb, struct lookup_intent *it)
2510 struct ll_sb_info *sbi = NULL;
2511 struct lustre_md md = { NULL };
2512 bool default_lmv_deleted = false;
2517 LASSERT(*inode || sb);
2518 sbi = sb ? ll_s2sbi(sb) : ll_i2sbi(*inode);
2519 rc = md_get_lustre_md(sbi->ll_md_exp, req, sbi->ll_dt_exp,
2520 sbi->ll_md_exp, &md);
2525 * clear default_lmv only if intent_getattr reply doesn't contain it.
2526 * but it needs to be done after iget, check this early because
2527 * ll_update_lsm_md() may change md.
2529 if (it && (it->it_op & (IT_LOOKUP | IT_GETATTR)) &&
2530 S_ISDIR(md.body->mbo_mode) && !md.default_lmv)
2531 default_lmv_deleted = true;
2534 rc = ll_update_inode(*inode, &md);
2538 LASSERT(sb != NULL);
2541 * At this point server returns to client's same fid as client
2542 * generated for creating. So using ->fid1 is okay here.
2544 if (!fid_is_sane(&md.body->mbo_fid1)) {
2545 CERROR("%s: Fid is insane "DFID"\n",
2547 PFID(&md.body->mbo_fid1));
2548 GOTO(out, rc = -EINVAL);
2551 *inode = ll_iget(sb, cl_fid_build_ino(&md.body->mbo_fid1,
2552 sbi->ll_flags & LL_SBI_32BIT_API),
2554 if (IS_ERR(*inode)) {
2555 #ifdef CONFIG_LUSTRE_FS_POSIX_ACL
2557 posix_acl_release(md.posix_acl);
2558 md.posix_acl = NULL;
2561 rc = IS_ERR(*inode) ? PTR_ERR(*inode) : -ENOMEM;
2563 CERROR("new_inode -fatal: rc %d\n", rc);
2568 /* Handling piggyback layout lock.
2569 * Layout lock can be piggybacked by getattr and open request.
2570 * The lsm can be applied to inode only if it comes with a layout lock
2571 * otherwise correct layout may be overwritten, for example:
2572 * 1. proc1: mdt returns a lsm but not granting layout
2573 * 2. layout was changed by another client
2574 * 3. proc2: refresh layout and layout lock granted
2575 * 4. proc1: to apply a stale layout */
2576 if (it != NULL && it->it_lock_mode != 0) {
2577 struct lustre_handle lockh;
2578 struct ldlm_lock *lock;
2580 lockh.cookie = it->it_lock_handle;
2581 lock = ldlm_handle2lock(&lockh);
2582 LASSERT(lock != NULL);
2583 if (ldlm_has_layout(lock)) {
2584 struct cl_object_conf conf;
2586 memset(&conf, 0, sizeof(conf));
2587 conf.coc_opc = OBJECT_CONF_SET;
2588 conf.coc_inode = *inode;
2589 conf.coc_lock = lock;
2590 conf.u.coc_layout = md.layout;
2591 (void)ll_layout_conf(*inode, &conf);
2593 LDLM_LOCK_PUT(lock);
2596 if (default_lmv_deleted)
2597 ll_update_default_lsm_md(*inode, &md);
2602 /* cleanup will be done if necessary */
2603 md_free_lustre_md(sbi->ll_md_exp, &md);
2605 if (rc != 0 && it != NULL && it->it_op & IT_OPEN)
2606 ll_open_cleanup(sb != NULL ? sb : (*inode)->i_sb, req);
2611 int ll_obd_statfs(struct inode *inode, void __user *arg)
2613 struct ll_sb_info *sbi = NULL;
2614 struct obd_export *exp;
2616 struct obd_ioctl_data *data = NULL;
2620 if (!inode || !(sbi = ll_i2sbi(inode)))
2621 GOTO(out_statfs, rc = -EINVAL);
2623 rc = obd_ioctl_getdata(&buf, &len, arg);
2625 GOTO(out_statfs, rc);
2628 if (!data->ioc_inlbuf1 || !data->ioc_inlbuf2 ||
2629 !data->ioc_pbuf1 || !data->ioc_pbuf2)
2630 GOTO(out_statfs, rc = -EINVAL);
2632 if (data->ioc_inllen1 != sizeof(__u32) ||
2633 data->ioc_inllen2 != sizeof(__u32) ||
2634 data->ioc_plen1 != sizeof(struct obd_statfs) ||
2635 data->ioc_plen2 != sizeof(struct obd_uuid))
2636 GOTO(out_statfs, rc = -EINVAL);
2638 memcpy(&type, data->ioc_inlbuf1, sizeof(__u32));
2639 if (type & LL_STATFS_LMV)
2640 exp = sbi->ll_md_exp;
2641 else if (type & LL_STATFS_LOV)
2642 exp = sbi->ll_dt_exp;
2644 GOTO(out_statfs, rc = -ENODEV);
2646 rc = obd_iocontrol(IOC_OBD_STATFS, exp, len, buf, NULL);
2648 GOTO(out_statfs, rc);
2650 OBD_FREE_LARGE(buf, len);
2655 * this is normally called in ll_fini_md_op_data(), but sometimes it needs to
2656 * be called early to avoid deadlock.
2658 void ll_unlock_md_op_lsm(struct md_op_data *op_data)
2660 if (op_data->op_mea2_sem) {
2661 up_read(op_data->op_mea2_sem);
2662 op_data->op_mea2_sem = NULL;
2665 if (op_data->op_mea1_sem) {
2666 up_read(op_data->op_mea1_sem);
2667 op_data->op_mea1_sem = NULL;
2671 /* this function prepares md_op_data hint for passing it down to MD stack. */
2672 struct md_op_data *ll_prep_md_op_data(struct md_op_data *op_data,
2673 struct inode *i1, struct inode *i2,
2674 const char *name, size_t namelen,
2675 __u32 mode, enum md_op_code opc,
2678 LASSERT(i1 != NULL);
2681 /* Do not reuse namelen for something else. */
2683 return ERR_PTR(-EINVAL);
2685 if (namelen > ll_i2sbi(i1)->ll_namelen)
2686 return ERR_PTR(-ENAMETOOLONG);
2688 if (!lu_name_is_valid_2(name, namelen))
2689 return ERR_PTR(-EINVAL);
2692 if (op_data == NULL)
2693 OBD_ALLOC_PTR(op_data);
2695 if (op_data == NULL)
2696 return ERR_PTR(-ENOMEM);
2698 ll_i2gids(op_data->op_suppgids, i1, i2);
2699 op_data->op_fid1 = *ll_inode2fid(i1);
2700 op_data->op_code = opc;
2702 if (S_ISDIR(i1->i_mode)) {
2703 down_read(&ll_i2info(i1)->lli_lsm_sem);
2704 op_data->op_mea1_sem = &ll_i2info(i1)->lli_lsm_sem;
2705 op_data->op_mea1 = ll_i2info(i1)->lli_lsm_md;
2706 op_data->op_default_mea1 = ll_i2info(i1)->lli_default_lsm_md;
2710 op_data->op_fid2 = *ll_inode2fid(i2);
2711 if (S_ISDIR(i2->i_mode)) {
2713 down_read(&ll_i2info(i2)->lli_lsm_sem);
2714 op_data->op_mea2_sem =
2715 &ll_i2info(i2)->lli_lsm_sem;
2717 op_data->op_mea2 = ll_i2info(i2)->lli_lsm_md;
2720 fid_zero(&op_data->op_fid2);
2723 if (ll_i2sbi(i1)->ll_flags & LL_SBI_64BIT_HASH)
2724 op_data->op_cli_flags |= CLI_HASH64;
2726 if (ll_need_32bit_api(ll_i2sbi(i1)))
2727 op_data->op_cli_flags |= CLI_API32;
2729 op_data->op_name = name;
2730 op_data->op_namelen = namelen;
2731 op_data->op_mode = mode;
2732 op_data->op_mod_time = ktime_get_real_seconds();
2733 op_data->op_fsuid = from_kuid(&init_user_ns, current_fsuid());
2734 op_data->op_fsgid = from_kgid(&init_user_ns, current_fsgid());
2735 op_data->op_cap = cfs_curproc_cap_pack();
2736 op_data->op_mds = 0;
2737 if ((opc == LUSTRE_OPC_CREATE) && (name != NULL) &&
2738 filename_is_volatile(name, namelen, &op_data->op_mds)) {
2739 op_data->op_bias |= MDS_CREATE_VOLATILE;
2741 op_data->op_data = data;
2746 void ll_finish_md_op_data(struct md_op_data *op_data)
2748 ll_unlock_md_op_lsm(op_data);
2749 security_release_secctx(op_data->op_file_secctx,
2750 op_data->op_file_secctx_size);
2751 OBD_FREE_PTR(op_data);
2754 #ifdef HAVE_SUPEROPS_USE_DENTRY
2755 int ll_show_options(struct seq_file *seq, struct dentry *dentry)
2757 int ll_show_options(struct seq_file *seq, struct vfsmount *vfs)
2760 struct ll_sb_info *sbi;
2762 #ifdef HAVE_SUPEROPS_USE_DENTRY
2763 LASSERT((seq != NULL) && (dentry != NULL));
2764 sbi = ll_s2sbi(dentry->d_sb);
2766 LASSERT((seq != NULL) && (vfs != NULL));
2767 sbi = ll_s2sbi(vfs->mnt_sb);
2770 if (sbi->ll_flags & LL_SBI_NOLCK)
2771 seq_puts(seq, ",nolock");
2773 if (sbi->ll_flags & LL_SBI_FLOCK)
2774 seq_puts(seq, ",flock");
2776 if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
2777 seq_puts(seq, ",localflock");
2779 if (sbi->ll_flags & LL_SBI_USER_XATTR)
2780 seq_puts(seq, ",user_xattr");
2782 if (sbi->ll_flags & LL_SBI_LAZYSTATFS)
2783 seq_puts(seq, ",lazystatfs");
2785 if (sbi->ll_flags & LL_SBI_USER_FID2PATH)
2786 seq_puts(seq, ",user_fid2path");
2788 if (sbi->ll_flags & LL_SBI_ALWAYS_PING)
2789 seq_puts(seq, ",always_ping");
2795 * Get obd name by cmd, and copy out to user space
2797 int ll_get_obd_name(struct inode *inode, unsigned int cmd, unsigned long arg)
2799 struct ll_sb_info *sbi = ll_i2sbi(inode);
2800 struct obd_device *obd;
2803 if (cmd == OBD_IOC_GETDTNAME)
2804 obd = class_exp2obd(sbi->ll_dt_exp);
2805 else if (cmd == OBD_IOC_GETMDNAME)
2806 obd = class_exp2obd(sbi->ll_md_exp);
2813 if (copy_to_user((void __user *)arg, obd->obd_name,
2814 strlen(obd->obd_name) + 1))
2820 static char* ll_d_path(struct dentry *dentry, char *buf, int bufsize)
2827 p.mnt = current->fs->root.mnt;
2829 path = d_path(&p, buf, bufsize);
2834 void ll_dirty_page_discard_warn(struct page *page, int ioret)
2836 char *buf, *path = NULL;
2837 struct dentry *dentry = NULL;
2838 struct inode *inode = page->mapping->host;
2840 /* this can be called inside spin lock so use GFP_ATOMIC. */
2841 buf = (char *)__get_free_page(GFP_ATOMIC);
2843 dentry = d_find_alias(page->mapping->host);
2845 path = ll_d_path(dentry, buf, PAGE_SIZE);
2848 /* The below message is checked in recovery-small.sh test_24b */
2850 "%s: dirty page discard: %s/fid: "DFID"/%s may get corrupted "
2851 "(rc %d)\n", ll_i2sbi(inode)->ll_fsname,
2852 s2lsi(page->mapping->host->i_sb)->lsi_lmd->lmd_dev,
2853 PFID(ll_inode2fid(inode)),
2854 (path && !IS_ERR(path)) ? path : "", ioret);
2860 free_page((unsigned long)buf);
2863 ssize_t ll_copy_user_md(const struct lov_user_md __user *md,
2864 struct lov_user_md **kbuf)
2866 struct lov_user_md lum;
2870 if (copy_from_user(&lum, md, sizeof(lum)))
2873 lum_size = ll_lov_user_md_size(&lum);
2877 OBD_ALLOC_LARGE(*kbuf, lum_size);
2881 if (copy_from_user(*kbuf, md, lum_size) != 0) {
2882 OBD_FREE_LARGE(*kbuf, lum_size);
2890 * Compute llite root squash state after a change of root squash
2891 * configuration setting or add/remove of a lnet nid
2893 void ll_compute_rootsquash_state(struct ll_sb_info *sbi)
2895 struct root_squash_info *squash = &sbi->ll_squash;
2898 struct lnet_process_id id;
2900 /* Update norootsquash flag */
2901 spin_lock(&squash->rsi_lock);
2902 if (list_empty(&squash->rsi_nosquash_nids))
2903 sbi->ll_flags &= ~LL_SBI_NOROOTSQUASH;
2905 /* Do not apply root squash as soon as one of our NIDs is
2906 * in the nosquash_nids list */
2909 while (LNetGetId(i++, &id) != -ENOENT) {
2910 if (LNET_NETTYP(LNET_NIDNET(id.nid)) == LOLND)
2912 if (cfs_match_nid(id.nid, &squash->rsi_nosquash_nids)) {
2918 sbi->ll_flags |= LL_SBI_NOROOTSQUASH;
2920 sbi->ll_flags &= ~LL_SBI_NOROOTSQUASH;
2922 spin_unlock(&squash->rsi_lock);
2926 * Parse linkea content to extract information about a given hardlink
2928 * \param[in] ldata - Initialized linkea data
2929 * \param[in] linkno - Link identifier
2930 * \param[out] parent_fid - The entry's parent FID
2931 * \param[out] ln - Entry name destination buffer
2933 * \retval 0 on success
2934 * \retval Appropriate negative error code on failure
2936 static int ll_linkea_decode(struct linkea_data *ldata, unsigned int linkno,
2937 struct lu_fid *parent_fid, struct lu_name *ln)
2943 rc = linkea_init_with_rec(ldata);
2947 if (linkno >= ldata->ld_leh->leh_reccount)
2948 /* beyond last link */
2951 linkea_first_entry(ldata);
2952 for (idx = 0; ldata->ld_lee != NULL; idx++) {
2953 linkea_entry_unpack(ldata->ld_lee, &ldata->ld_reclen, ln,
2958 linkea_next_entry(ldata);
2968 * Get parent FID and name of an identified link. Operation is performed for
2969 * a given link number, letting the caller iterate over linkno to list one or
2970 * all links of an entry.
2972 * \param[in] file - File descriptor against which to perform the operation
2973 * \param[in,out] arg - User-filled structure containing the linkno to operate
2974 * on and the available size. It is eventually filled with
2975 * the requested information or left untouched on error
2977 * \retval - 0 on success
2978 * \retval - Appropriate negative error code on failure
2980 int ll_getparent(struct file *file, struct getparent __user *arg)
2982 struct inode *inode = file_inode(file);
2983 struct linkea_data *ldata;
2984 struct lu_buf buf = LU_BUF_NULL;
2986 struct lu_fid parent_fid;
2993 if (!cfs_capable(CFS_CAP_DAC_READ_SEARCH) &&
2994 !(ll_i2sbi(inode)->ll_flags & LL_SBI_USER_FID2PATH))
2997 if (get_user(name_size, &arg->gp_name_size))
3000 if (get_user(linkno, &arg->gp_linkno))
3003 if (name_size > PATH_MAX)
3006 OBD_ALLOC(ldata, sizeof(*ldata));
3010 rc = linkea_data_new(ldata, &buf);
3012 GOTO(ldata_free, rc);
3014 rc = ll_xattr_list(inode, XATTR_NAME_LINK, XATTR_TRUSTED_T, buf.lb_buf,
3015 buf.lb_len, OBD_MD_FLXATTR);
3019 rc = ll_linkea_decode(ldata, linkno, &parent_fid, &ln);
3023 if (ln.ln_namelen >= name_size)
3024 GOTO(lb_free, rc = -EOVERFLOW);
3026 if (copy_to_user(&arg->gp_fid, &parent_fid, sizeof(arg->gp_fid)))
3027 GOTO(lb_free, rc = -EFAULT);
3029 if (copy_to_user(&arg->gp_name, ln.ln_name, ln.ln_namelen))
3030 GOTO(lb_free, rc = -EFAULT);
3032 if (put_user('\0', arg->gp_name + ln.ln_namelen))
3033 GOTO(lb_free, rc = -EFAULT);
3038 OBD_FREE(ldata, sizeof(*ldata));