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/statfs.h>
41 #include <linux/time.h>
42 #include <linux/types.h>
43 #include <linux/version.h>
45 #include <linux/user_namespace.h>
46 #ifdef HAVE_UIDGID_HEADER
47 # include <linux/uidgid.h>
49 #include <linux/security.h>
51 #include <uapi/linux/lustre/lustre_ioctl.h>
52 #ifdef HAVE_UAPI_LINUX_MOUNT_H
53 #include <uapi/linux/mount.h>
56 #include <lustre_ha.h>
57 #include <lustre_dlm.h>
58 #include <lprocfs_status.h>
59 #include <lustre_disk.h>
60 #include <uapi/linux/lustre/lustre_param.h>
61 #include <lustre_log.h>
62 #include <cl_object.h>
63 #include <obd_cksum.h>
64 #include "llite_internal.h"
66 struct kmem_cache *ll_file_data_slab;
69 #define log2(n) ffz(~(n))
72 static struct ll_sb_info *ll_init_sbi(void)
74 struct ll_sb_info *sbi = NULL;
76 unsigned long lru_page_max;
86 spin_lock_init(&sbi->ll_lock);
87 mutex_init(&sbi->ll_lco.lco_lock);
88 spin_lock_init(&sbi->ll_pp_extent_lock);
89 spin_lock_init(&sbi->ll_process_lock);
90 sbi->ll_rw_stats_on = 0;
91 sbi->ll_statfs_max_age = OBD_STATFS_CACHE_SECONDS;
94 pages = si.totalram - si.totalhigh;
95 lru_page_max = pages / 2;
97 /* initialize ll_cache data */
98 sbi->ll_cache = cl_cache_init(lru_page_max);
99 if (sbi->ll_cache == NULL) {
100 OBD_FREE(sbi, sizeof(*sbi));
104 sbi->ll_ra_info.ra_max_pages_per_file = min(pages / 32,
105 SBI_DEFAULT_READAHEAD_MAX);
106 sbi->ll_ra_info.ra_max_pages = sbi->ll_ra_info.ra_max_pages_per_file;
107 sbi->ll_ra_info.ra_max_read_ahead_whole_pages = -1;
109 ll_generate_random_uuid(uuid);
110 class_uuid_unparse(uuid, &sbi->ll_sb_uuid);
111 CDEBUG(D_CONFIG, "generated uuid: %s\n", sbi->ll_sb_uuid.uuid);
113 sbi->ll_flags |= LL_SBI_VERBOSE;
114 #ifdef ENABLE_CHECKSUM
115 sbi->ll_flags |= LL_SBI_CHECKSUM;
118 sbi->ll_flags |= LL_SBI_FLOCK;
121 #ifdef HAVE_LRU_RESIZE_SUPPORT
122 sbi->ll_flags |= LL_SBI_LRU_RESIZE;
124 sbi->ll_flags |= LL_SBI_LAZYSTATFS;
126 for (i = 0; i <= LL_PROCESS_HIST_MAX; i++) {
127 spin_lock_init(&sbi->ll_rw_extents_info.pp_extents[i].
129 spin_lock_init(&sbi->ll_rw_extents_info.pp_extents[i].
133 /* metadata statahead is enabled by default */
134 sbi->ll_sa_running_max = LL_SA_RUNNING_DEF;
135 sbi->ll_sa_max = LL_SA_RPC_DEF;
136 atomic_set(&sbi->ll_sa_total, 0);
137 atomic_set(&sbi->ll_sa_wrong, 0);
138 atomic_set(&sbi->ll_sa_running, 0);
139 atomic_set(&sbi->ll_agl_total, 0);
140 sbi->ll_flags |= LL_SBI_AGL_ENABLED;
141 sbi->ll_flags |= LL_SBI_FAST_READ;
142 sbi->ll_flags |= LL_SBI_TINY_WRITE;
145 sbi->ll_squash.rsi_uid = 0;
146 sbi->ll_squash.rsi_gid = 0;
147 INIT_LIST_HEAD(&sbi->ll_squash.rsi_nosquash_nids);
148 init_rwsem(&sbi->ll_squash.rsi_sem);
153 static void ll_free_sbi(struct super_block *sb)
155 struct ll_sb_info *sbi = ll_s2sbi(sb);
159 if (!list_empty(&sbi->ll_squash.rsi_nosquash_nids))
160 cfs_free_nidlist(&sbi->ll_squash.rsi_nosquash_nids);
161 if (sbi->ll_cache != NULL) {
162 cl_cache_decref(sbi->ll_cache);
163 sbi->ll_cache = NULL;
165 OBD_FREE(sbi, sizeof(*sbi));
170 static int client_common_fill_super(struct super_block *sb, char *md, char *dt,
171 struct vfsmount *mnt)
173 struct inode *root = NULL;
174 struct ll_sb_info *sbi = ll_s2sbi(sb);
175 struct obd_statfs *osfs = NULL;
176 struct ptlrpc_request *request = NULL;
177 struct obd_connect_data *data = NULL;
178 struct obd_uuid *uuid;
179 struct md_op_data *op_data;
180 struct lustre_md lmd;
182 int size, err, checksum;
185 sbi->ll_md_obd = class_name2obd(md);
186 if (!sbi->ll_md_obd) {
187 CERROR("MD %s: not setup or attached\n", md);
201 /* pass client page size via ocd_grant_blkbits, the server should report
202 * back its backend blocksize for grant calculation purpose */
203 data->ocd_grant_blkbits = PAGE_SHIFT;
205 /* indicate MDT features supported by this client */
206 data->ocd_connect_flags = OBD_CONNECT_IBITS | OBD_CONNECT_NODEVOH |
207 OBD_CONNECT_ATTRFID | OBD_CONNECT_GRANT |
208 OBD_CONNECT_VERSION | OBD_CONNECT_BRW_SIZE |
209 OBD_CONNECT_SRVLOCK | OBD_CONNECT_TRUNCLOCK|
210 OBD_CONNECT_MDS_CAPA | OBD_CONNECT_OSS_CAPA |
211 OBD_CONNECT_CANCELSET | OBD_CONNECT_FID |
212 OBD_CONNECT_AT | OBD_CONNECT_LOV_V3 |
213 OBD_CONNECT_VBR | OBD_CONNECT_FULL20 |
214 OBD_CONNECT_64BITHASH |
215 OBD_CONNECT_EINPROGRESS |
216 OBD_CONNECT_JOBSTATS | OBD_CONNECT_LVB_TYPE |
217 OBD_CONNECT_LAYOUTLOCK | OBD_CONNECT_PINGLESS|
218 OBD_CONNECT_MAX_EASIZE |
219 OBD_CONNECT_FLOCK_DEAD |
220 OBD_CONNECT_DISP_STRIPE | OBD_CONNECT_LFSCK |
221 OBD_CONNECT_OPEN_BY_FID |
222 OBD_CONNECT_DIR_STRIPE |
223 OBD_CONNECT_BULK_MBITS | OBD_CONNECT_CKSUM |
224 OBD_CONNECT_SUBTREE |
225 OBD_CONNECT_MULTIMODRPCS |
226 OBD_CONNECT_GRANT_PARAM |
227 OBD_CONNECT_SHORTIO | OBD_CONNECT_FLAGS2;
229 data->ocd_connect_flags2 = OBD_CONNECT2_FLR |
230 OBD_CONNECT2_LOCK_CONVERT |
231 OBD_CONNECT2_DIR_MIGRATE |
232 OBD_CONNECT2_SUM_STATFS |
233 OBD_CONNECT2_ARCHIVE_ID_ARRAY |
235 OBD_CONNECT2_ASYNC_DISCARD |
236 OBD_CONNECT2_GETATTR_PFID;
238 #ifdef HAVE_LRU_RESIZE_SUPPORT
239 if (sbi->ll_flags & LL_SBI_LRU_RESIZE)
240 data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
242 #ifdef CONFIG_FS_POSIX_ACL
243 data->ocd_connect_flags |= OBD_CONNECT_ACL | OBD_CONNECT_UMASK |
244 OBD_CONNECT_LARGE_ACL;
247 data->ocd_cksum_types = obd_cksum_types_supported_client();
249 if (OBD_FAIL_CHECK(OBD_FAIL_MDC_LIGHTWEIGHT))
250 /* flag mdc connection as lightweight, only used for test
251 * purpose, use with care */
252 data->ocd_connect_flags |= OBD_CONNECT_LIGHTWEIGHT;
254 data->ocd_ibits_known = MDS_INODELOCK_FULL;
255 data->ocd_version = LUSTRE_VERSION_CODE;
257 if (sb->s_flags & SB_RDONLY)
258 data->ocd_connect_flags |= OBD_CONNECT_RDONLY;
259 if (sbi->ll_flags & LL_SBI_USER_XATTR)
260 data->ocd_connect_flags |= OBD_CONNECT_XATTR;
263 /* Setting this indicates we correctly support S_NOSEC (See kernel
264 * commit 9e1f1de02c2275d7172e18dc4e7c2065777611bf)
266 sb->s_flags |= SB_NOSEC;
269 if (sbi->ll_flags & LL_SBI_FLOCK)
270 sbi->ll_fop = &ll_file_operations_flock;
271 else if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
272 sbi->ll_fop = &ll_file_operations;
274 sbi->ll_fop = &ll_file_operations_noflock;
276 /* always ping even if server suppress_pings */
277 if (sbi->ll_flags & LL_SBI_ALWAYS_PING)
278 data->ocd_connect_flags &= ~OBD_CONNECT_PINGLESS;
280 obd_connect_set_secctx(data);
282 #if defined(CONFIG_SECURITY)
283 data->ocd_connect_flags2 |= OBD_CONNECT2_SELINUX_POLICY;
286 data->ocd_brw_size = MD_MAX_BRW_SIZE;
288 err = obd_connect(NULL, &sbi->ll_md_exp, sbi->ll_md_obd,
289 &sbi->ll_sb_uuid, data, NULL);
291 LCONSOLE_ERROR_MSG(0x14f, "An MDT (md %s) is performing "
292 "recovery, of which this client is not a "
293 "part. Please wait for recovery to complete,"
294 " abort, or time out.\n", md);
297 CERROR("cannot connect to %s: rc = %d\n", md, err);
301 sbi->ll_md_exp->exp_connect_data = *data;
303 err = obd_fid_init(sbi->ll_md_exp->exp_obd, sbi->ll_md_exp,
304 LUSTRE_SEQ_METADATA);
306 CERROR("%s: Can't init metadata layer FID infrastructure, "
307 "rc = %d\n", sbi->ll_md_exp->exp_obd->obd_name, err);
311 /* For mount, we only need fs info from MDT0, and also in DNE, it
312 * can make sure the client can be mounted as long as MDT0 is
314 err = obd_statfs(NULL, sbi->ll_md_exp, osfs,
315 ktime_get_seconds() - sbi->ll_statfs_max_age,
316 OBD_STATFS_FOR_MDT0);
318 GOTO(out_md_fid, err);
320 /* This needs to be after statfs to ensure connect has finished.
321 * Note that "data" does NOT contain the valid connect reply.
322 * If connecting to a 1.8 server there will be no LMV device, so
323 * we can access the MDC export directly and exp_connect_flags will
324 * be non-zero, but if accessing an upgraded 2.1 server it will
325 * have the correct flags filled in.
326 * XXX: fill in the LMV exp_connect_flags from MDC(s). */
327 valid = exp_connect_flags(sbi->ll_md_exp) & CLIENT_CONNECT_MDT_REQD;
328 if (exp_connect_flags(sbi->ll_md_exp) != 0 &&
329 valid != CLIENT_CONNECT_MDT_REQD) {
332 OBD_ALLOC_WAIT(buf, PAGE_SIZE);
333 obd_connect_flags2str(buf, PAGE_SIZE,
334 valid ^ CLIENT_CONNECT_MDT_REQD, 0, ",");
335 LCONSOLE_ERROR_MSG(0x170, "Server %s does not support "
336 "feature(s) needed for correct operation "
337 "of this client (%s). Please upgrade "
338 "server or downgrade client.\n",
339 sbi->ll_md_exp->exp_obd->obd_name, buf);
340 OBD_FREE(buf, PAGE_SIZE);
341 GOTO(out_md_fid, err = -EPROTO);
344 size = sizeof(*data);
345 err = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_CONN_DATA),
346 KEY_CONN_DATA, &size, data);
348 CERROR("%s: Get connect data failed: rc = %d\n",
349 sbi->ll_md_exp->exp_obd->obd_name, err);
350 GOTO(out_md_fid, err);
353 LASSERT(osfs->os_bsize);
354 sb->s_blocksize = osfs->os_bsize;
355 sb->s_blocksize_bits = log2(osfs->os_bsize);
356 sb->s_magic = LL_SUPER_MAGIC;
357 sb->s_maxbytes = MAX_LFS_FILESIZE;
358 sbi->ll_namelen = osfs->os_namelen;
359 sbi->ll_mnt.mnt = current->fs->root.mnt;
361 if ((sbi->ll_flags & LL_SBI_USER_XATTR) &&
362 !(data->ocd_connect_flags & OBD_CONNECT_XATTR)) {
363 LCONSOLE_INFO("Disabling user_xattr feature because "
364 "it is not supported on the server\n");
365 sbi->ll_flags &= ~LL_SBI_USER_XATTR;
368 if (data->ocd_connect_flags & OBD_CONNECT_ACL) {
370 sb->s_flags |= SB_POSIXACL;
372 sbi->ll_flags |= LL_SBI_ACL;
374 LCONSOLE_INFO("client wants to enable acl, but mdt not!\n");
376 sb->s_flags &= ~SB_POSIXACL;
378 sbi->ll_flags &= ~LL_SBI_ACL;
381 if (data->ocd_connect_flags & OBD_CONNECT_64BITHASH)
382 sbi->ll_flags |= LL_SBI_64BIT_HASH;
384 if (data->ocd_connect_flags & OBD_CONNECT_LAYOUTLOCK)
385 sbi->ll_flags |= LL_SBI_LAYOUT_LOCK;
387 if (obd_connect_has_secctx(data))
388 sbi->ll_flags |= LL_SBI_FILE_SECCTX;
390 if (data->ocd_ibits_known & MDS_INODELOCK_XATTR) {
391 if (!(data->ocd_connect_flags & OBD_CONNECT_MAX_EASIZE)) {
392 LCONSOLE_INFO("%s: disabling xattr cache due to "
393 "unknown maximum xattr size.\n", dt);
394 } else if (!sbi->ll_xattr_cache_set) {
395 /* If xattr_cache is already set (no matter 0 or 1)
396 * during processing llog, it won't be enabled here. */
397 sbi->ll_flags |= LL_SBI_XATTR_CACHE;
398 sbi->ll_xattr_cache_enabled = 1;
402 sbi->ll_dt_obd = class_name2obd(dt);
403 if (!sbi->ll_dt_obd) {
404 CERROR("DT %s: not setup or attached\n", dt);
405 GOTO(out_md_fid, err = -ENODEV);
408 /* pass client page size via ocd_grant_blkbits, the server should report
409 * back its backend blocksize for grant calculation purpose */
410 data->ocd_grant_blkbits = PAGE_SHIFT;
412 /* indicate OST features supported by this client */
413 data->ocd_connect_flags = OBD_CONNECT_GRANT | OBD_CONNECT_VERSION |
414 OBD_CONNECT_REQPORTAL | OBD_CONNECT_BRW_SIZE |
415 OBD_CONNECT_CANCELSET | OBD_CONNECT_FID |
416 OBD_CONNECT_SRVLOCK | OBD_CONNECT_TRUNCLOCK|
417 OBD_CONNECT_AT | OBD_CONNECT_OSS_CAPA |
418 OBD_CONNECT_VBR | OBD_CONNECT_FULL20 |
419 OBD_CONNECT_64BITHASH | OBD_CONNECT_MAXBYTES |
420 OBD_CONNECT_EINPROGRESS |
421 OBD_CONNECT_JOBSTATS | OBD_CONNECT_LVB_TYPE |
422 OBD_CONNECT_LAYOUTLOCK |
423 OBD_CONNECT_PINGLESS | OBD_CONNECT_LFSCK |
424 OBD_CONNECT_BULK_MBITS | OBD_CONNECT_SHORTIO |
425 OBD_CONNECT_FLAGS2 | OBD_CONNECT_GRANT_SHRINK;
427 /* The client currently advertises support for OBD_CONNECT_LOCKAHEAD_OLD so it
428 * can interoperate with an older version of lockahead which was released prior
429 * to landing in master. This support will be dropped when 2.13 development
430 * starts. At the point, we should not just drop the connect flag (below), we
431 * should also remove the support in the code.
433 * Removing it means a few things:
434 * 1. Remove this section here
435 * 2. Remove CEF_NONBLOCK in ll_file_lockahead()
436 * 3. Remove function exp_connect_lockahead_old
437 * 4. Remove LDLM_FL_LOCKAHEAD_OLD_RESERVED in lustre_dlm_flags.h
439 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 12, 50, 0)
440 data->ocd_connect_flags |= OBD_CONNECT_LOCKAHEAD_OLD;
443 data->ocd_connect_flags2 = OBD_CONNECT2_LOCKAHEAD;
445 if (!OBD_FAIL_CHECK(OBD_FAIL_OSC_CONNECT_GRANT_PARAM))
446 data->ocd_connect_flags |= OBD_CONNECT_GRANT_PARAM;
448 /* OBD_CONNECT_CKSUM should always be set, even if checksums are
449 * disabled by default, because it can still be enabled on the
450 * fly via /sys. As a consequence, we still need to come to an
451 * agreement on the supported algorithms at connect time
453 data->ocd_connect_flags |= OBD_CONNECT_CKSUM;
455 if (OBD_FAIL_CHECK(OBD_FAIL_OSC_CKSUM_ADLER_ONLY))
456 data->ocd_cksum_types = OBD_CKSUM_ADLER;
458 data->ocd_cksum_types = obd_cksum_types_supported_client();
460 #ifdef HAVE_LRU_RESIZE_SUPPORT
461 data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
463 /* always ping even if server suppress_pings */
464 if (sbi->ll_flags & LL_SBI_ALWAYS_PING)
465 data->ocd_connect_flags &= ~OBD_CONNECT_PINGLESS;
467 CDEBUG(D_RPCTRACE, "ocd_connect_flags: %#llx ocd_version: %d "
468 "ocd_grant: %d\n", data->ocd_connect_flags,
469 data->ocd_version, data->ocd_grant);
471 sbi->ll_dt_obd->obd_upcall.onu_owner = &sbi->ll_lco;
472 sbi->ll_dt_obd->obd_upcall.onu_upcall = cl_ocd_update;
474 data->ocd_brw_size = DT_MAX_BRW_SIZE;
476 err = obd_connect(NULL, &sbi->ll_dt_exp, sbi->ll_dt_obd,
477 &sbi->ll_sb_uuid, data, NULL);
479 LCONSOLE_ERROR_MSG(0x150, "An OST (dt %s) is performing "
480 "recovery, of which this client is not a "
481 "part. Please wait for recovery to "
482 "complete, abort, or time out.\n", dt);
485 CERROR("%s: Cannot connect to %s: rc = %d\n",
486 sbi->ll_dt_exp->exp_obd->obd_name, dt, err);
490 sbi->ll_dt_exp->exp_connect_data = *data;
492 /* Don't change value if it was specified in the config log */
493 if (sbi->ll_ra_info.ra_max_read_ahead_whole_pages == -1) {
494 sbi->ll_ra_info.ra_max_read_ahead_whole_pages =
495 max_t(unsigned long, SBI_DEFAULT_READAHEAD_WHOLE_MAX,
496 (data->ocd_brw_size >> PAGE_SHIFT));
497 if (sbi->ll_ra_info.ra_max_read_ahead_whole_pages >
498 sbi->ll_ra_info.ra_max_pages_per_file)
499 sbi->ll_ra_info.ra_max_read_ahead_whole_pages =
500 sbi->ll_ra_info.ra_max_pages_per_file;
503 err = obd_fid_init(sbi->ll_dt_exp->exp_obd, sbi->ll_dt_exp,
504 LUSTRE_SEQ_METADATA);
506 CERROR("%s: Can't init data layer FID infrastructure, "
507 "rc = %d\n", sbi->ll_dt_exp->exp_obd->obd_name, err);
511 mutex_lock(&sbi->ll_lco.lco_lock);
512 sbi->ll_lco.lco_flags = data->ocd_connect_flags;
513 sbi->ll_lco.lco_md_exp = sbi->ll_md_exp;
514 sbi->ll_lco.lco_dt_exp = sbi->ll_dt_exp;
515 mutex_unlock(&sbi->ll_lco.lco_lock);
517 fid_zero(&sbi->ll_root_fid);
518 err = md_get_root(sbi->ll_md_exp, get_mount_fileset(sb),
521 CERROR("cannot mds_connect: rc = %d\n", err);
522 GOTO(out_lock_cn_cb, err);
524 if (!fid_is_sane(&sbi->ll_root_fid)) {
525 CERROR("%s: Invalid root fid "DFID" during mount\n",
526 sbi->ll_md_exp->exp_obd->obd_name,
527 PFID(&sbi->ll_root_fid));
528 GOTO(out_lock_cn_cb, err = -EINVAL);
530 CDEBUG(D_SUPER, "rootfid "DFID"\n", PFID(&sbi->ll_root_fid));
532 sb->s_op = &lustre_super_operations;
533 #ifdef HAVE_XATTR_HANDLER_FLAGS
534 sb->s_xattr = ll_xattr_handlers;
536 #if THREAD_SIZE >= 8192 /*b=17630*/
537 sb->s_export_op = &lustre_export_operations;
541 * XXX: move this to after cbd setup? */
542 valid = OBD_MD_FLGETATTR | OBD_MD_FLBLOCKS | OBD_MD_FLMODEASIZE;
543 if (sbi->ll_flags & LL_SBI_ACL)
544 valid |= OBD_MD_FLACL;
546 OBD_ALLOC_PTR(op_data);
548 GOTO(out_lock_cn_cb, err = -ENOMEM);
550 op_data->op_fid1 = sbi->ll_root_fid;
551 op_data->op_mode = 0;
552 op_data->op_valid = valid;
554 err = md_getattr(sbi->ll_md_exp, op_data, &request);
556 OBD_FREE_PTR(op_data);
558 CERROR("%s: md_getattr failed for root: rc = %d\n",
559 sbi->ll_md_exp->exp_obd->obd_name, err);
560 GOTO(out_lock_cn_cb, err);
563 err = md_get_lustre_md(sbi->ll_md_exp, request, sbi->ll_dt_exp,
564 sbi->ll_md_exp, &lmd);
566 CERROR("failed to understand root inode md: rc = %d\n", err);
567 ptlrpc_req_finished(request);
568 GOTO(out_lock_cn_cb, err);
571 LASSERT(fid_is_sane(&sbi->ll_root_fid));
572 root = ll_iget(sb, cl_fid_build_ino(&sbi->ll_root_fid,
573 sbi->ll_flags & LL_SBI_32BIT_API),
575 md_free_lustre_md(sbi->ll_md_exp, &lmd);
576 ptlrpc_req_finished(request);
579 #ifdef CONFIG_FS_POSIX_ACL
581 posix_acl_release(lmd.posix_acl);
582 lmd.posix_acl = NULL;
585 err = IS_ERR(root) ? PTR_ERR(root) : -EBADF;
587 CERROR("lustre_lite: bad iget4 for root\n");
591 checksum = sbi->ll_flags & LL_SBI_CHECKSUM;
592 if (sbi->ll_checksum_set) {
593 err = obd_set_info_async(NULL, sbi->ll_dt_exp,
594 sizeof(KEY_CHECKSUM), KEY_CHECKSUM,
595 sizeof(checksum), &checksum, NULL);
597 CERROR("%s: Set checksum failed: rc = %d\n",
598 sbi->ll_dt_exp->exp_obd->obd_name, err);
604 err = obd_set_info_async(NULL, sbi->ll_dt_exp, sizeof(KEY_CACHE_SET),
605 KEY_CACHE_SET, sizeof(*sbi->ll_cache),
606 sbi->ll_cache, NULL);
608 CERROR("%s: Set cache_set failed: rc = %d\n",
609 sbi->ll_dt_exp->exp_obd->obd_name, err);
613 sb->s_root = d_make_root(root);
614 if (sb->s_root == NULL) {
615 CERROR("%s: can't make root dentry\n",
616 ll_get_fsname(sb, NULL, 0));
617 GOTO(out_root, err = -ENOMEM);
619 #ifdef HAVE_DCACHE_LOCK
620 sb->s_root->d_op = &ll_d_ops;
623 sbi->ll_sdev_orig = sb->s_dev;
625 /* We set sb->s_dev equal on all lustre clients in order to support
626 * NFS export clustering. NFSD requires that the FSID be the same
628 /* s_dev is also used in lt_compare() to compare two fs, but that is
629 * only a node-local comparison. */
630 uuid = obd_get_uuid(sbi->ll_md_exp);
632 sb->s_dev = get_uuid2int(uuid->uuid, strlen(uuid->uuid));
639 if (sbi->ll_dt_obd) {
640 err = sysfs_create_link(&sbi->ll_kset.kobj,
641 &sbi->ll_dt_obd->obd_kset.kobj,
642 sbi->ll_dt_obd->obd_type->typ_name);
644 CERROR("%s: could not register %s in llite: rc = %d\n",
645 dt, ll_get_fsname(sb, NULL, 0), err);
650 if (sbi->ll_md_obd) {
651 err = sysfs_create_link(&sbi->ll_kset.kobj,
652 &sbi->ll_md_obd->obd_kset.kobj,
653 sbi->ll_md_obd->obd_type->typ_name);
655 CERROR("%s: could not register %s in llite: rc = %d\n",
656 md, ll_get_fsname(sb, NULL, 0), err);
666 obd_fid_fini(sbi->ll_dt_exp->exp_obd);
668 obd_disconnect(sbi->ll_dt_exp);
669 sbi->ll_dt_exp = NULL;
670 sbi->ll_dt_obd = NULL;
672 obd_fid_fini(sbi->ll_md_exp->exp_obd);
674 obd_disconnect(sbi->ll_md_exp);
675 sbi->ll_md_exp = NULL;
676 sbi->ll_md_obd = NULL;
685 int ll_get_max_mdsize(struct ll_sb_info *sbi, int *lmmsize)
689 size = sizeof(*lmmsize);
690 rc = obd_get_info(NULL, sbi->ll_dt_exp, sizeof(KEY_MAX_EASIZE),
691 KEY_MAX_EASIZE, &size, lmmsize);
693 CERROR("%s: cannot get max LOV EA size: rc = %d\n",
694 sbi->ll_dt_exp->exp_obd->obd_name, rc);
699 rc = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_MAX_EASIZE),
700 KEY_MAX_EASIZE, &size, lmmsize);
702 CERROR("Get max mdsize error rc %d\n", rc);
708 * Get the value of the default_easize parameter.
710 * \see client_obd::cl_default_mds_easize
712 * \param[in] sbi superblock info for this filesystem
713 * \param[out] lmmsize pointer to storage location for value
715 * \retval 0 on success
716 * \retval negative negated errno on failure
718 int ll_get_default_mdsize(struct ll_sb_info *sbi, int *lmmsize)
723 rc = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_DEFAULT_EASIZE),
724 KEY_DEFAULT_EASIZE, &size, lmmsize);
726 CERROR("Get default mdsize error rc %d\n", rc);
732 * Set the default_easize parameter to the given value.
734 * \see client_obd::cl_default_mds_easize
736 * \param[in] sbi superblock info for this filesystem
737 * \param[in] lmmsize the size to set
739 * \retval 0 on success
740 * \retval negative negated errno on failure
742 int ll_set_default_mdsize(struct ll_sb_info *sbi, int lmmsize)
746 if (lmmsize < sizeof(struct lov_mds_md) ||
747 lmmsize > OBD_MAX_DEFAULT_EA_SIZE)
750 rc = obd_set_info_async(NULL, sbi->ll_md_exp,
751 sizeof(KEY_DEFAULT_EASIZE), KEY_DEFAULT_EASIZE,
752 sizeof(int), &lmmsize, NULL);
757 static void client_common_put_super(struct super_block *sb)
759 struct ll_sb_info *sbi = ll_s2sbi(sb);
764 obd_fid_fini(sbi->ll_dt_exp->exp_obd);
765 obd_disconnect(sbi->ll_dt_exp);
766 sbi->ll_dt_exp = NULL;
768 ll_debugfs_unregister_super(sb);
770 obd_fid_fini(sbi->ll_md_exp->exp_obd);
771 obd_disconnect(sbi->ll_md_exp);
772 sbi->ll_md_exp = NULL;
777 void ll_kill_super(struct super_block *sb)
779 struct ll_sb_info *sbi;
783 if (!(sb->s_flags & SB_ACTIVE))
787 /* we need restore s_dev from changed for clustred NFS before put_super
788 * because new kernels have cached s_dev and change sb->s_dev in
789 * put_super not affected real removing devices */
791 sb->s_dev = sbi->ll_sdev_orig;
792 sbi->ll_umounting = 1;
794 /* wait running statahead threads to quit */
795 while (atomic_read(&sbi->ll_sa_running) > 0) {
796 set_current_state(TASK_UNINTERRUPTIBLE);
797 schedule_timeout(msecs_to_jiffies(MSEC_PER_SEC >> 3));
804 static inline int ll_set_opt(const char *opt, char *data, int fl)
806 if (strncmp(opt, data, strlen(opt)) != 0)
812 /* non-client-specific mount options are parsed in lmd_parse */
813 static int ll_options(char *options, struct ll_sb_info *sbi)
816 char *s1 = options, *s2;
817 int *flags = &sbi->ll_flags;
823 CDEBUG(D_CONFIG, "Parsing opts %s\n", options);
826 CDEBUG(D_SUPER, "next opt=%s\n", s1);
827 tmp = ll_set_opt("nolock", s1, LL_SBI_NOLCK);
832 tmp = ll_set_opt("flock", s1, LL_SBI_FLOCK);
834 *flags = (*flags & ~LL_SBI_LOCALFLOCK) | tmp;
837 tmp = ll_set_opt("localflock", s1, LL_SBI_LOCALFLOCK);
839 *flags = (*flags & ~LL_SBI_FLOCK) | tmp;
842 tmp = ll_set_opt("noflock", s1, LL_SBI_FLOCK|LL_SBI_LOCALFLOCK);
847 tmp = ll_set_opt("user_xattr", s1, LL_SBI_USER_XATTR);
852 tmp = ll_set_opt("nouser_xattr", s1, LL_SBI_USER_XATTR);
857 tmp = ll_set_opt("context", s1, 1);
860 tmp = ll_set_opt("fscontext", s1, 1);
863 tmp = ll_set_opt("defcontext", s1, 1);
866 tmp = ll_set_opt("rootcontext", s1, 1);
869 tmp = ll_set_opt("user_fid2path", s1, LL_SBI_USER_FID2PATH);
874 tmp = ll_set_opt("nouser_fid2path", s1, LL_SBI_USER_FID2PATH);
880 tmp = ll_set_opt("checksum", s1, LL_SBI_CHECKSUM);
883 sbi->ll_checksum_set = 1;
886 tmp = ll_set_opt("nochecksum", s1, LL_SBI_CHECKSUM);
889 sbi->ll_checksum_set = 1;
892 tmp = ll_set_opt("lruresize", s1, LL_SBI_LRU_RESIZE);
897 tmp = ll_set_opt("nolruresize", s1, LL_SBI_LRU_RESIZE);
902 tmp = ll_set_opt("lazystatfs", s1, LL_SBI_LAZYSTATFS);
907 tmp = ll_set_opt("nolazystatfs", s1, LL_SBI_LAZYSTATFS);
912 tmp = ll_set_opt("32bitapi", s1, LL_SBI_32BIT_API);
917 tmp = ll_set_opt("verbose", s1, LL_SBI_VERBOSE);
922 tmp = ll_set_opt("noverbose", s1, LL_SBI_VERBOSE);
927 tmp = ll_set_opt("always_ping", s1, LL_SBI_ALWAYS_PING);
932 LCONSOLE_ERROR_MSG(0x152, "Unknown option '%s', won't mount.\n",
938 s2 = strchr(s1, ',');
946 void ll_lli_init(struct ll_inode_info *lli)
948 lli->lli_inode_magic = LLI_INODE_MAGIC;
950 spin_lock_init(&lli->lli_lock);
951 lli->lli_posix_acl = NULL;
952 /* Do not set lli_fid, it has been initialized already. */
953 fid_zero(&lli->lli_pfid);
954 lli->lli_mds_read_och = NULL;
955 lli->lli_mds_write_och = NULL;
956 lli->lli_mds_exec_och = NULL;
957 lli->lli_open_fd_read_count = 0;
958 lli->lli_open_fd_write_count = 0;
959 lli->lli_open_fd_exec_count = 0;
960 mutex_init(&lli->lli_och_mutex);
961 spin_lock_init(&lli->lli_agl_lock);
962 spin_lock_init(&lli->lli_layout_lock);
963 ll_layout_version_set(lli, CL_LAYOUT_GEN_NONE);
964 lli->lli_clob = NULL;
966 init_rwsem(&lli->lli_xattrs_list_rwsem);
967 mutex_init(&lli->lli_xattrs_enq_lock);
969 LASSERT(lli->lli_vfs_inode.i_mode != 0);
970 if (S_ISDIR(lli->lli_vfs_inode.i_mode)) {
971 mutex_init(&lli->lli_readdir_mutex);
972 lli->lli_opendir_key = NULL;
974 spin_lock_init(&lli->lli_sa_lock);
975 lli->lli_opendir_pid = 0;
976 lli->lli_sa_enabled = 0;
977 lli->lli_def_stripe_offset = -1;
978 init_rwsem(&lli->lli_lsm_sem);
980 mutex_init(&lli->lli_size_mutex);
981 lli->lli_symlink_name = NULL;
982 init_rwsem(&lli->lli_trunc_sem);
983 range_lock_tree_init(&lli->lli_write_tree);
984 init_rwsem(&lli->lli_glimpse_sem);
985 lli->lli_glimpse_time = ktime_set(0, 0);
986 INIT_LIST_HEAD(&lli->lli_agl_list);
987 lli->lli_agl_index = 0;
988 lli->lli_async_rc = 0;
990 mutex_init(&lli->lli_layout_mutex);
991 memset(lli->lli_jobid, 0, sizeof(lli->lli_jobid));
994 #define MAX_STRING_SIZE 128
996 #ifndef HAVE_SUPER_SETUP_BDI_NAME
998 #define LSI_BDI_INITIALIZED 0x00400000
1000 #ifndef HAVE_BDI_CAP_MAP_COPY
1001 # define BDI_CAP_MAP_COPY 0
1004 static int super_setup_bdi_name(struct super_block *sb, char *fmt, ...)
1006 struct lustre_sb_info *lsi = s2lsi(sb);
1007 char buf[MAX_STRING_SIZE];
1011 err = bdi_init(&lsi->lsi_bdi);
1015 lsi->lsi_flags |= LSI_BDI_INITIALIZED;
1016 lsi->lsi_bdi.capabilities = BDI_CAP_MAP_COPY;
1017 lsi->lsi_bdi.name = "lustre";
1018 va_start(args, fmt);
1019 vsnprintf(buf, MAX_STRING_SIZE, fmt, args);
1021 err = bdi_register(&lsi->lsi_bdi, NULL, "%s", buf);
1024 sb->s_bdi = &lsi->lsi_bdi;
1028 #endif /* !HAVE_SUPER_SETUP_BDI_NAME */
1030 int ll_fill_super(struct super_block *sb, struct vfsmount *mnt)
1032 struct lustre_profile *lprof = NULL;
1033 struct lustre_sb_info *lsi = s2lsi(sb);
1034 struct ll_sb_info *sbi = NULL;
1035 char *dt = NULL, *md = NULL;
1036 char *profilenm = get_profile_name(sb);
1037 struct config_llog_instance *cfg;
1038 /* %p for void* in printf needs 16+2 characters: 0xffffffffffffffff */
1039 const int instlen = 16 + 2;
1040 unsigned long cfg_instance = ll_get_cfg_instance(sb);
1041 char name[MAX_STRING_SIZE];
1049 /* for ASLR, to map between cfg_instance and hashed ptr */
1050 CDEBUG(D_VFSTRACE, "VFS Op: cfg_instance %s-%016lx (sb %p)\n",
1051 profilenm, cfg_instance, sb);
1053 try_module_get(THIS_MODULE);
1057 GOTO(out_free_cfg, err = -ENOMEM);
1059 /* client additional sb info */
1060 lsi->lsi_llsbi = sbi = ll_init_sbi();
1062 GOTO(out_free_cfg, err = -ENOMEM);
1064 err = ll_options(lsi->lsi_lmd->lmd_opts, sbi);
1066 GOTO(out_free_cfg, err);
1068 err = super_setup_bdi_name(sb, "lustre-%016lx", cfg_instance);
1070 GOTO(out_free_cfg, err);
1072 #ifndef HAVE_DCACHE_LOCK
1073 /* kernel >= 2.6.38 store dentry operations in sb->s_d_op. */
1074 sb->s_d_op = &ll_d_ops;
1077 len = strlen(profilenm);
1078 ptr = strrchr(profilenm, '-');
1079 if (ptr && (strcmp(ptr, "-client") == 0))
1083 snprintf(name, MAX_STRING_SIZE, "%.*s-%016lx", len,
1084 profilenm, cfg_instance);
1086 /* Call ll_debugfs_register_super() before lustre_process_log()
1087 * so that "llite.*.*" params can be processed correctly.
1089 err = ll_debugfs_register_super(sb, name);
1091 CERROR("%s: could not register mountpoint in llite: rc = %d\n",
1092 ll_get_fsname(sb, NULL, 0), err);
1096 /* The cfg_instance is a value unique to this super, in case some
1097 * joker tries to mount the same fs at two mount points.
1099 cfg->cfg_instance = cfg_instance;
1100 cfg->cfg_uuid = lsi->lsi_llsbi->ll_sb_uuid;
1101 cfg->cfg_callback = class_config_llog_handler;
1102 cfg->cfg_sub_clds = CONFIG_SUB_CLIENT;
1103 /* set up client obds */
1104 err = lustre_process_log(sb, profilenm, cfg);
1106 GOTO(out_debugfs, err);
1108 /* Profile set with LCFG_MOUNTOPT so we can find our mdc and osc obds */
1109 lprof = class_get_profile(profilenm);
1110 if (lprof == NULL) {
1111 LCONSOLE_ERROR_MSG(0x156, "The client profile '%s' could not be"
1112 " read from the MGS. Does that filesystem "
1113 "exist?\n", profilenm);
1114 GOTO(out_debugfs, err = -EINVAL);
1116 CDEBUG(D_CONFIG, "Found profile %s: mdc=%s osc=%s\n", profilenm,
1117 lprof->lp_md, lprof->lp_dt);
1119 dt_len = strlen(lprof->lp_dt) + instlen + 2;
1120 OBD_ALLOC(dt, dt_len);
1122 GOTO(out_profile, err = -ENOMEM);
1123 snprintf(dt, dt_len - 1, "%s-%016lx", lprof->lp_dt, cfg_instance);
1125 md_len = strlen(lprof->lp_md) + instlen + 2;
1126 OBD_ALLOC(md, md_len);
1128 GOTO(out_free_dt, err = -ENOMEM);
1129 snprintf(md, md_len - 1, "%s-%016lx", lprof->lp_md, cfg_instance);
1131 /* connections, registrations, sb setup */
1132 err = client_common_fill_super(sb, md, dt, mnt);
1134 GOTO(out_free_md, err);
1136 sbi->ll_client_common_fill_super_succeeded = 1;
1140 OBD_FREE(md, md_len);
1143 OBD_FREE(dt, dt_len);
1146 class_put_profile(lprof);
1149 ll_debugfs_unregister_super(sb);
1156 else if (sbi->ll_flags & LL_SBI_VERBOSE)
1157 LCONSOLE_WARN("Mounted %s\n", profilenm);
1159 } /* ll_fill_super */
1161 void ll_put_super(struct super_block *sb)
1163 struct config_llog_instance cfg, params_cfg;
1164 struct obd_device *obd;
1165 struct lustre_sb_info *lsi = s2lsi(sb);
1166 struct ll_sb_info *sbi = ll_s2sbi(sb);
1167 char *profilenm = get_profile_name(sb);
1168 unsigned long cfg_instance = ll_get_cfg_instance(sb);
1170 int next, force = 1, rc = 0;
1174 GOTO(out_no_sbi, 0);
1176 /* Should replace instance_id with something better for ASLR */
1177 CDEBUG(D_VFSTRACE, "VFS Op: cfg_instance %s-%016lx (sb %p)\n",
1178 profilenm, cfg_instance, sb);
1180 cfg.cfg_instance = cfg_instance;
1181 lustre_end_log(sb, profilenm, &cfg);
1183 params_cfg.cfg_instance = cfg_instance;
1184 lustre_end_log(sb, PARAMS_FILENAME, ¶ms_cfg);
1186 if (sbi->ll_md_exp) {
1187 obd = class_exp2obd(sbi->ll_md_exp);
1189 force = obd->obd_force;
1192 /* Wait for unstable pages to be committed to stable storage */
1194 struct l_wait_info lwi = LWI_INTR(LWI_ON_SIGNAL_NOOP, NULL);
1195 rc = l_wait_event(sbi->ll_cache->ccc_unstable_waitq,
1196 atomic_long_read(&sbi->ll_cache->ccc_unstable_nr) == 0,
1200 ccc_count = atomic_long_read(&sbi->ll_cache->ccc_unstable_nr);
1201 if (force == 0 && rc != -EINTR)
1202 LASSERTF(ccc_count == 0, "count: %li\n", ccc_count);
1204 /* We need to set force before the lov_disconnect in
1205 lustre_common_put_super, since l_d cleans up osc's as well. */
1208 while ((obd = class_devices_in_group(&sbi->ll_sb_uuid,
1210 obd->obd_force = force;
1214 if (sbi->ll_client_common_fill_super_succeeded) {
1215 /* Only if client_common_fill_super succeeded */
1216 client_common_put_super(sb);
1220 while ((obd = class_devices_in_group(&sbi->ll_sb_uuid, &next)) !=NULL) {
1221 class_manual_cleanup(obd);
1224 if (sbi->ll_flags & LL_SBI_VERBOSE)
1225 LCONSOLE_WARN("Unmounted %s\n", profilenm ? profilenm : "");
1228 class_del_profile(profilenm);
1230 #ifndef HAVE_SUPER_SETUP_BDI_NAME
1231 if (lsi->lsi_flags & LSI_BDI_INITIALIZED) {
1232 bdi_destroy(&lsi->lsi_bdi);
1233 lsi->lsi_flags &= ~LSI_BDI_INITIALIZED;
1238 lsi->lsi_llsbi = NULL;
1240 lustre_common_put_super(sb);
1242 cl_env_cache_purge(~0);
1244 module_put(THIS_MODULE);
1247 } /* client_put_super */
1249 struct inode *ll_inode_from_resource_lock(struct ldlm_lock *lock)
1251 struct inode *inode = NULL;
1253 /* NOTE: we depend on atomic igrab() -bzzz */
1254 lock_res_and_lock(lock);
1255 if (lock->l_resource->lr_lvb_inode) {
1256 struct ll_inode_info * lli;
1257 lli = ll_i2info(lock->l_resource->lr_lvb_inode);
1258 if (lli->lli_inode_magic == LLI_INODE_MAGIC) {
1259 inode = igrab(lock->l_resource->lr_lvb_inode);
1261 inode = lock->l_resource->lr_lvb_inode;
1262 LDLM_DEBUG_LIMIT(inode->i_state & I_FREEING ? D_INFO :
1263 D_WARNING, lock, "lr_lvb_inode %p is "
1264 "bogus: magic %08x",
1265 lock->l_resource->lr_lvb_inode,
1266 lli->lli_inode_magic);
1270 unlock_res_and_lock(lock);
1274 void ll_dir_clear_lsm_md(struct inode *inode)
1276 struct ll_inode_info *lli = ll_i2info(inode);
1278 LASSERT(S_ISDIR(inode->i_mode));
1280 if (lli->lli_lsm_md != NULL) {
1281 lmv_free_memmd(lli->lli_lsm_md);
1282 lli->lli_lsm_md = NULL;
1286 static struct inode *ll_iget_anon_dir(struct super_block *sb,
1287 const struct lu_fid *fid,
1288 struct lustre_md *md)
1290 struct ll_sb_info *sbi = ll_s2sbi(sb);
1291 struct mdt_body *body = md->body;
1292 struct inode *inode;
1296 ino = cl_fid_build_ino(fid, sbi->ll_flags & LL_SBI_32BIT_API);
1297 inode = iget_locked(sb, ino);
1298 if (inode == NULL) {
1299 CERROR("%s: failed get simple inode "DFID": rc = -ENOENT\n",
1300 ll_get_fsname(sb, NULL, 0), PFID(fid));
1301 RETURN(ERR_PTR(-ENOENT));
1304 if (inode->i_state & I_NEW) {
1305 struct ll_inode_info *lli = ll_i2info(inode);
1306 struct lmv_stripe_md *lsm = md->lmv;
1308 inode->i_mode = (inode->i_mode & ~S_IFMT) |
1309 (body->mbo_mode & S_IFMT);
1310 LASSERTF(S_ISDIR(inode->i_mode), "Not slave inode "DFID"\n",
1313 inode->i_mtime.tv_sec = 0;
1314 inode->i_atime.tv_sec = 0;
1315 inode->i_ctime.tv_sec = 0;
1318 #ifdef HAVE_BACKING_DEV_INFO
1319 /* initializing backing dev info. */
1320 inode->i_mapping->backing_dev_info =
1321 &s2lsi(inode->i_sb)->lsi_bdi;
1323 inode->i_op = &ll_dir_inode_operations;
1324 inode->i_fop = &ll_dir_operations;
1325 lli->lli_fid = *fid;
1328 LASSERT(lsm != NULL);
1329 /* master object FID */
1330 lli->lli_pfid = body->mbo_fid1;
1331 CDEBUG(D_INODE, "lli %p slave "DFID" master "DFID"\n",
1332 lli, PFID(fid), PFID(&lli->lli_pfid));
1333 unlock_new_inode(inode);
1339 static int ll_init_lsm_md(struct inode *inode, struct lustre_md *md)
1342 struct lmv_stripe_md *lsm = md->lmv;
1343 struct ll_inode_info *lli = ll_i2info(inode);
1346 LASSERT(lsm != NULL);
1348 CDEBUG(D_INODE, "%s: "DFID" set dir layout:\n",
1349 ll_get_fsname(inode->i_sb, NULL, 0),
1350 PFID(&lli->lli_fid));
1351 lsm_md_dump(D_INODE, lsm);
1353 /* XXX sigh, this lsm_root initialization should be in
1354 * LMV layer, but it needs ll_iget right now, so we
1355 * put this here right now. */
1356 for (i = 0; i < lsm->lsm_md_stripe_count; i++) {
1357 fid = &lsm->lsm_md_oinfo[i].lmo_fid;
1358 LASSERT(lsm->lsm_md_oinfo[i].lmo_root == NULL);
1360 if (!fid_is_sane(fid))
1363 /* Unfortunately ll_iget will call ll_update_inode,
1364 * where the initialization of slave inode is slightly
1365 * different, so it reset lsm_md to NULL to avoid
1366 * initializing lsm for slave inode. */
1367 lsm->lsm_md_oinfo[i].lmo_root =
1368 ll_iget_anon_dir(inode->i_sb, fid, md);
1369 if (IS_ERR(lsm->lsm_md_oinfo[i].lmo_root)) {
1370 int rc = PTR_ERR(lsm->lsm_md_oinfo[i].lmo_root);
1372 lsm->lsm_md_oinfo[i].lmo_root = NULL;
1374 iput(lsm->lsm_md_oinfo[i].lmo_root);
1375 lsm->lsm_md_oinfo[i].lmo_root = NULL;
1381 lli->lli_lsm_md = lsm;
1386 static int ll_update_lsm_md(struct inode *inode, struct lustre_md *md)
1388 struct ll_inode_info *lli = ll_i2info(inode);
1389 struct lmv_stripe_md *lsm = md->lmv;
1390 struct cl_attr *attr;
1395 LASSERT(S_ISDIR(inode->i_mode));
1396 CDEBUG(D_INODE, "update lsm %p of "DFID"\n", lli->lli_lsm_md,
1397 PFID(ll_inode2fid(inode)));
1400 * no striped information from request, lustre_md from req does not
1401 * include stripeEA, see ll_md_setattr()
1407 * normally dir layout doesn't change, only take read lock to check
1408 * that to avoid blocking other MD operations.
1410 down_read(&lli->lli_lsm_sem);
1412 /* some concurrent lookup initialized lsm, and unchanged */
1413 if (lli->lli_lsm_md && lsm_md_eq(lli->lli_lsm_md, lsm))
1414 GOTO(unlock, rc = 0);
1416 /* if dir layout doesn't match, check whether version is increased,
1417 * which means layout is changed, this happens in dir split/merge and
1420 if (lli->lli_lsm_md &&
1421 lsm->lsm_md_layout_version <=
1422 lli->lli_lsm_md->lsm_md_layout_version) {
1423 CERROR("%s: "DFID" dir layout mismatch:\n",
1424 ll_get_fsname(inode->i_sb, NULL, 0),
1425 PFID(&lli->lli_fid));
1426 lsm_md_dump(D_ERROR, lli->lli_lsm_md);
1427 lsm_md_dump(D_ERROR, lsm);
1428 GOTO(unlock, rc = -EINVAL);
1431 up_read(&lli->lli_lsm_sem);
1432 down_write(&lli->lli_lsm_sem);
1433 /* clear existing lsm */
1434 if (lli->lli_lsm_md) {
1435 lmv_free_memmd(lli->lli_lsm_md);
1436 lli->lli_lsm_md = NULL;
1439 rc = ll_init_lsm_md(inode, md);
1440 up_write(&lli->lli_lsm_sem);
1444 /* set md->lmv to NULL, so the following free lustre_md will not free
1449 /* md_merge_attr() may take long, since lsm is already set, switch to
1452 down_read(&lli->lli_lsm_sem);
1453 OBD_ALLOC_PTR(attr);
1455 GOTO(unlock, rc = -ENOMEM);
1457 /* validate the lsm */
1458 rc = md_merge_attr(ll_i2mdexp(inode), lli->lli_lsm_md, attr,
1459 ll_md_blocking_ast);
1461 if (md->body->mbo_valid & OBD_MD_FLNLINK)
1462 md->body->mbo_nlink = attr->cat_nlink;
1463 if (md->body->mbo_valid & OBD_MD_FLSIZE)
1464 md->body->mbo_size = attr->cat_size;
1465 if (md->body->mbo_valid & OBD_MD_FLATIME)
1466 md->body->mbo_atime = attr->cat_atime;
1467 if (md->body->mbo_valid & OBD_MD_FLCTIME)
1468 md->body->mbo_ctime = attr->cat_ctime;
1469 if (md->body->mbo_valid & OBD_MD_FLMTIME)
1470 md->body->mbo_mtime = attr->cat_mtime;
1476 up_read(&lli->lli_lsm_sem);
1481 void ll_clear_inode(struct inode *inode)
1483 struct ll_inode_info *lli = ll_i2info(inode);
1484 struct ll_sb_info *sbi = ll_i2sbi(inode);
1487 CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p)\n",
1488 PFID(ll_inode2fid(inode)), inode);
1490 if (S_ISDIR(inode->i_mode)) {
1491 /* these should have been cleared in ll_file_release */
1492 LASSERT(lli->lli_opendir_key == NULL);
1493 LASSERT(lli->lli_sai == NULL);
1494 LASSERT(lli->lli_opendir_pid == 0);
1497 md_null_inode(sbi->ll_md_exp, ll_inode2fid(inode));
1499 LASSERT(!lli->lli_open_fd_write_count);
1500 LASSERT(!lli->lli_open_fd_read_count);
1501 LASSERT(!lli->lli_open_fd_exec_count);
1503 if (lli->lli_mds_write_och)
1504 ll_md_real_close(inode, FMODE_WRITE);
1505 if (lli->lli_mds_exec_och)
1506 ll_md_real_close(inode, FMODE_EXEC);
1507 if (lli->lli_mds_read_och)
1508 ll_md_real_close(inode, FMODE_READ);
1510 if (S_ISLNK(inode->i_mode) && lli->lli_symlink_name) {
1511 OBD_FREE(lli->lli_symlink_name,
1512 strlen(lli->lli_symlink_name) + 1);
1513 lli->lli_symlink_name = NULL;
1516 ll_xattr_cache_destroy(inode);
1518 #ifdef CONFIG_FS_POSIX_ACL
1519 forget_all_cached_acls(inode);
1520 if (lli->lli_posix_acl) {
1521 posix_acl_release(lli->lli_posix_acl);
1522 lli->lli_posix_acl = NULL;
1525 lli->lli_inode_magic = LLI_INODE_DEAD;
1527 if (S_ISDIR(inode->i_mode))
1528 ll_dir_clear_lsm_md(inode);
1529 else if (S_ISREG(inode->i_mode) && !is_bad_inode(inode))
1530 LASSERT(list_empty(&lli->lli_agl_list));
1533 * XXX This has to be done before lsm is freed below, because
1534 * cl_object still uses inode lsm.
1536 cl_inode_fini(inode);
1541 static int ll_md_setattr(struct dentry *dentry, struct md_op_data *op_data)
1543 struct lustre_md md;
1544 struct inode *inode = dentry->d_inode;
1545 struct ll_sb_info *sbi = ll_i2sbi(inode);
1546 struct ptlrpc_request *request = NULL;
1550 op_data = ll_prep_md_op_data(op_data, inode, NULL, NULL, 0, 0,
1551 LUSTRE_OPC_ANY, NULL);
1552 if (IS_ERR(op_data))
1553 RETURN(PTR_ERR(op_data));
1555 rc = md_setattr(sbi->ll_md_exp, op_data, NULL, 0, &request);
1557 ptlrpc_req_finished(request);
1558 if (rc == -ENOENT) {
1560 /* Unlinked special device node? Or just a race?
1561 * Pretend we done everything. */
1562 if (!S_ISREG(inode->i_mode) &&
1563 !S_ISDIR(inode->i_mode)) {
1564 ia_valid = op_data->op_attr.ia_valid;
1565 op_data->op_attr.ia_valid &= ~TIMES_SET_FLAGS;
1566 rc = simple_setattr(dentry, &op_data->op_attr);
1567 op_data->op_attr.ia_valid = ia_valid;
1569 } else if (rc != -EPERM && rc != -EACCES && rc != -ETXTBSY) {
1570 CERROR("md_setattr fails: rc = %d\n", rc);
1575 rc = md_get_lustre_md(sbi->ll_md_exp, request, sbi->ll_dt_exp,
1576 sbi->ll_md_exp, &md);
1578 ptlrpc_req_finished(request);
1582 ia_valid = op_data->op_attr.ia_valid;
1583 /* inode size will be in ll_setattr_ost, can't do it now since dirty
1584 * cache is not cleared yet. */
1585 op_data->op_attr.ia_valid &= ~(TIMES_SET_FLAGS | ATTR_SIZE);
1586 if (S_ISREG(inode->i_mode))
1588 rc = simple_setattr(dentry, &op_data->op_attr);
1589 if (S_ISREG(inode->i_mode))
1590 inode_unlock(inode);
1591 op_data->op_attr.ia_valid = ia_valid;
1593 rc = ll_update_inode(inode, &md);
1594 ptlrpc_req_finished(request);
1599 /* If this inode has objects allocated to it (lsm != NULL), then the OST
1600 * object(s) determine the file size and mtime. Otherwise, the MDS will
1601 * keep these values until such a time that objects are allocated for it.
1602 * We do the MDS operations first, as it is checking permissions for us.
1603 * We don't to the MDS RPC if there is nothing that we want to store there,
1604 * otherwise there is no harm in updating mtime/atime on the MDS if we are
1605 * going to do an RPC anyways.
1607 * If we are doing a truncate, we will send the mtime and ctime updates
1608 * to the OST with the punch RPC, otherwise we do an explicit setattr RPC.
1609 * I don't believe it is possible to get e.g. ATTR_MTIME_SET and ATTR_SIZE
1612 * In case of HSMimport, we only set attr on MDS.
1614 int ll_setattr_raw(struct dentry *dentry, struct iattr *attr,
1615 enum op_xvalid xvalid, bool hsm_import)
1617 struct inode *inode = dentry->d_inode;
1618 struct ll_inode_info *lli = ll_i2info(inode);
1619 struct md_op_data *op_data = NULL;
1623 CDEBUG(D_VFSTRACE, "%s: setattr inode "DFID"(%p) from %llu to %llu, "
1624 "valid %x, hsm_import %d\n",
1625 ll_get_fsname(inode->i_sb, NULL, 0), PFID(&lli->lli_fid),
1626 inode, i_size_read(inode), attr->ia_size, attr->ia_valid,
1629 if (attr->ia_valid & ATTR_SIZE) {
1630 /* Check new size against VFS/VM file size limit and rlimit */
1631 rc = inode_newsize_ok(inode, attr->ia_size);
1635 /* The maximum Lustre file size is variable, based on the
1636 * OST maximum object size and number of stripes. This
1637 * needs another check in addition to the VFS check above. */
1638 if (attr->ia_size > ll_file_maxbytes(inode)) {
1639 CDEBUG(D_INODE,"file "DFID" too large %llu > %llu\n",
1640 PFID(&lli->lli_fid), attr->ia_size,
1641 ll_file_maxbytes(inode));
1645 attr->ia_valid |= ATTR_MTIME | ATTR_CTIME;
1648 /* POSIX: check before ATTR_*TIME_SET set (from inode_change_ok) */
1649 if (attr->ia_valid & TIMES_SET_FLAGS) {
1650 if ((!uid_eq(current_fsuid(), inode->i_uid)) &&
1651 !cfs_capable(CFS_CAP_FOWNER))
1655 /* We mark all of the fields "set" so MDS/OST does not re-set them */
1656 if (!(xvalid & OP_XVALID_CTIME_SET) &&
1657 (attr->ia_valid & ATTR_CTIME)) {
1658 attr->ia_ctime = current_time(inode);
1659 xvalid |= OP_XVALID_CTIME_SET;
1661 if (!(attr->ia_valid & ATTR_ATIME_SET) &&
1662 (attr->ia_valid & ATTR_ATIME)) {
1663 attr->ia_atime = current_time(inode);
1664 attr->ia_valid |= ATTR_ATIME_SET;
1666 if (!(attr->ia_valid & ATTR_MTIME_SET) &&
1667 (attr->ia_valid & ATTR_MTIME)) {
1668 attr->ia_mtime = current_time(inode);
1669 attr->ia_valid |= ATTR_MTIME_SET;
1672 if (attr->ia_valid & (ATTR_MTIME | ATTR_CTIME))
1673 CDEBUG(D_INODE, "setting mtime %lld, ctime %lld, now = %lld\n",
1674 (s64)attr->ia_mtime.tv_sec, (s64)attr->ia_ctime.tv_sec,
1675 ktime_get_real_seconds());
1677 if (S_ISREG(inode->i_mode)) {
1678 if (attr->ia_valid & ATTR_SIZE)
1679 inode_dio_write_done(inode);
1680 inode_unlock(inode);
1683 /* We always do an MDS RPC, even if we're only changing the size;
1684 * only the MDS knows whether truncate() should fail with -ETXTBUSY */
1686 OBD_ALLOC_PTR(op_data);
1687 if (op_data == NULL)
1688 GOTO(out, rc = -ENOMEM);
1690 if (!hsm_import && attr->ia_valid & ATTR_SIZE) {
1691 /* If we are changing file size, file content is
1692 * modified, flag it.
1694 xvalid |= OP_XVALID_OWNEROVERRIDE;
1695 op_data->op_bias |= MDS_DATA_MODIFIED;
1696 ll_file_clear_flag(lli, LLIF_DATA_MODIFIED);
1699 if (attr->ia_valid & ATTR_FILE) {
1700 struct ll_file_data *fd = LUSTRE_FPRIVATE(attr->ia_file);
1702 if (fd->fd_lease_och)
1703 op_data->op_bias |= MDS_TRUNC_KEEP_LEASE;
1706 op_data->op_attr = *attr;
1707 op_data->op_xvalid = xvalid;
1709 rc = ll_md_setattr(dentry, op_data);
1713 if (!S_ISREG(inode->i_mode) || hsm_import)
1716 if (attr->ia_valid & (ATTR_SIZE | ATTR_ATIME | ATTR_ATIME_SET |
1717 ATTR_MTIME | ATTR_MTIME_SET | ATTR_CTIME) ||
1718 xvalid & OP_XVALID_CTIME_SET) {
1719 /* For truncate and utimes sending attributes to OSTs, setting
1720 * mtime/atime to the past will be performed under PW [0:EOF]
1721 * extent lock (new_size:EOF for truncate). It may seem
1722 * excessive to send mtime/atime updates to OSTs when not
1723 * setting times to past, but it is necessary due to possible
1724 * time de-synchronization between MDT inode and OST objects
1726 rc = cl_setattr_ost(lli->lli_clob, attr, xvalid, 0);
1729 /* If the file was restored, it needs to set dirty flag.
1731 * We've already sent MDS_DATA_MODIFIED flag in
1732 * ll_md_setattr() for truncate. However, the MDT refuses to
1733 * set the HS_DIRTY flag on released files, so we have to set
1734 * it again if the file has been restored. Please check how
1735 * LLIF_DATA_MODIFIED is set in vvp_io_setattr_fini().
1737 * Please notice that if the file is not released, the previous
1738 * MDS_DATA_MODIFIED has taken effect and usually
1739 * LLIF_DATA_MODIFIED is not set(see vvp_io_setattr_fini()).
1740 * This way we can save an RPC for common open + trunc
1742 if (ll_file_test_and_clear_flag(lli, LLIF_DATA_MODIFIED)) {
1743 struct hsm_state_set hss = {
1744 .hss_valid = HSS_SETMASK,
1745 .hss_setmask = HS_DIRTY,
1749 rc2 = ll_hsm_state_set(inode, &hss);
1750 /* truncate and write can happen at the same time, so that
1751 * the file can be set modified even though the file is not
1752 * restored from released state, and ll_hsm_state_set() is
1753 * not applicable for the file, and rc2 < 0 is normal in this
1756 CDEBUG(D_INFO, DFID "HSM set dirty failed: rc2 = %d\n",
1757 PFID(ll_inode2fid(inode)), rc2);
1762 if (op_data != NULL)
1763 ll_finish_md_op_data(op_data);
1765 if (S_ISREG(inode->i_mode)) {
1767 if ((attr->ia_valid & ATTR_SIZE) && !hsm_import)
1768 inode_dio_wait(inode);
1769 /* Once we've got the i_mutex, it's safe to set the S_NOSEC
1770 * flag. ll_update_inode (called from ll_md_setattr), clears
1771 * inode flags, so there is a gap where S_NOSEC is not set.
1772 * This can cause a writer to take the i_mutex unnecessarily,
1773 * but this is safe to do and should be rare. */
1774 inode_has_no_xattr(inode);
1777 ll_stats_ops_tally(ll_i2sbi(inode), (attr->ia_valid & ATTR_SIZE) ?
1778 LPROC_LL_TRUNC : LPROC_LL_SETATTR, 1);
1783 int ll_setattr(struct dentry *de, struct iattr *attr)
1785 int mode = de->d_inode->i_mode;
1786 enum op_xvalid xvalid = 0;
1788 if ((attr->ia_valid & (ATTR_CTIME|ATTR_SIZE|ATTR_MODE)) ==
1789 (ATTR_CTIME|ATTR_SIZE|ATTR_MODE))
1790 xvalid |= OP_XVALID_OWNEROVERRIDE;
1792 if (((attr->ia_valid & (ATTR_MODE|ATTR_FORCE|ATTR_SIZE)) ==
1793 (ATTR_SIZE|ATTR_MODE)) &&
1794 (((mode & S_ISUID) && !(attr->ia_mode & S_ISUID)) ||
1795 (((mode & (S_ISGID|S_IXGRP)) == (S_ISGID|S_IXGRP)) &&
1796 !(attr->ia_mode & S_ISGID))))
1797 attr->ia_valid |= ATTR_FORCE;
1799 if ((attr->ia_valid & ATTR_MODE) &&
1801 !(attr->ia_mode & S_ISUID) &&
1802 !(attr->ia_valid & ATTR_KILL_SUID))
1803 attr->ia_valid |= ATTR_KILL_SUID;
1805 if ((attr->ia_valid & ATTR_MODE) &&
1806 ((mode & (S_ISGID|S_IXGRP)) == (S_ISGID|S_IXGRP)) &&
1807 !(attr->ia_mode & S_ISGID) &&
1808 !(attr->ia_valid & ATTR_KILL_SGID))
1809 attr->ia_valid |= ATTR_KILL_SGID;
1811 return ll_setattr_raw(de, attr, xvalid, false);
1814 int ll_statfs_internal(struct ll_sb_info *sbi, struct obd_statfs *osfs,
1817 struct obd_statfs obd_osfs = { 0 };
1822 max_age = ktime_get_seconds() - sbi->ll_statfs_max_age;
1824 rc = obd_statfs(NULL, sbi->ll_md_exp, osfs, max_age, flags);
1828 osfs->os_type = LL_SUPER_MAGIC;
1830 CDEBUG(D_SUPER, "MDC blocks %llu/%llu objects %llu/%llu\n",
1831 osfs->os_bavail, osfs->os_blocks, osfs->os_ffree, osfs->os_files);
1833 if (osfs->os_state & OS_STATE_SUM)
1836 if (sbi->ll_flags & LL_SBI_LAZYSTATFS)
1837 flags |= OBD_STATFS_NODELAY;
1839 rc = obd_statfs(NULL, sbi->ll_dt_exp, &obd_osfs, max_age, flags);
1840 if (rc) /* Possibly a filesystem with no OSTs. Report MDT totals. */
1843 CDEBUG(D_SUPER, "OSC blocks %llu/%llu objects %llu/%llu\n",
1844 obd_osfs.os_bavail, obd_osfs.os_blocks, obd_osfs.os_ffree,
1847 osfs->os_bsize = obd_osfs.os_bsize;
1848 osfs->os_blocks = obd_osfs.os_blocks;
1849 osfs->os_bfree = obd_osfs.os_bfree;
1850 osfs->os_bavail = obd_osfs.os_bavail;
1852 /* If we have _some_ OSTs, but don't have as many free objects on the
1853 * OSTs as inodes on the MDTs, reduce the reported number of inodes
1854 * to compensate, so that the "inodes in use" number is correct.
1855 * This should be kept in sync with lod_statfs() behaviour.
1857 if (obd_osfs.os_files && obd_osfs.os_ffree < osfs->os_ffree) {
1858 osfs->os_files = (osfs->os_files - osfs->os_ffree) +
1860 osfs->os_ffree = obd_osfs.os_ffree;
1866 int ll_statfs(struct dentry *de, struct kstatfs *sfs)
1868 struct super_block *sb = de->d_sb;
1869 struct obd_statfs osfs;
1870 __u64 fsid = huge_encode_dev(sb->s_dev);
1873 CDEBUG(D_VFSTRACE, "VFS Op: at %llu jiffies\n", get_jiffies_64());
1874 ll_stats_ops_tally(ll_s2sbi(sb), LPROC_LL_STAFS, 1);
1876 /* Some amount of caching on the client is allowed */
1877 rc = ll_statfs_internal(ll_s2sbi(sb), &osfs, OBD_STATFS_SUM);
1881 statfs_unpack(sfs, &osfs);
1883 /* We need to downshift for all 32-bit kernels, because we can't
1884 * tell if the kernel is being called via sys_statfs64() or not.
1885 * Stop before overflowing f_bsize - in which case it is better
1886 * to just risk EOVERFLOW if caller is using old sys_statfs(). */
1887 if (sizeof(long) < 8) {
1888 while (osfs.os_blocks > ~0UL && sfs->f_bsize < 0x40000000) {
1891 osfs.os_blocks >>= 1;
1892 osfs.os_bfree >>= 1;
1893 osfs.os_bavail >>= 1;
1897 sfs->f_blocks = osfs.os_blocks;
1898 sfs->f_bfree = osfs.os_bfree;
1899 sfs->f_bavail = osfs.os_bavail;
1900 sfs->f_fsid.val[0] = (__u32)fsid;
1901 sfs->f_fsid.val[1] = (__u32)(fsid >> 32);
1905 void ll_inode_size_lock(struct inode *inode)
1907 struct ll_inode_info *lli;
1909 LASSERT(!S_ISDIR(inode->i_mode));
1911 lli = ll_i2info(inode);
1912 mutex_lock(&lli->lli_size_mutex);
1915 void ll_inode_size_unlock(struct inode *inode)
1917 struct ll_inode_info *lli;
1919 lli = ll_i2info(inode);
1920 mutex_unlock(&lli->lli_size_mutex);
1923 void ll_update_inode_flags(struct inode *inode, int ext_flags)
1925 inode->i_flags = ll_ext_to_inode_flags(ext_flags);
1926 if (ext_flags & LUSTRE_PROJINHERIT_FL)
1927 ll_file_set_flag(ll_i2info(inode), LLIF_PROJECT_INHERIT);
1929 ll_file_clear_flag(ll_i2info(inode), LLIF_PROJECT_INHERIT);
1932 int ll_update_inode(struct inode *inode, struct lustre_md *md)
1934 struct ll_inode_info *lli = ll_i2info(inode);
1935 struct mdt_body *body = md->body;
1936 struct ll_sb_info *sbi = ll_i2sbi(inode);
1939 if (body->mbo_valid & OBD_MD_FLEASIZE) {
1940 rc = cl_file_inode_init(inode, md);
1945 if (S_ISDIR(inode->i_mode)) {
1946 rc = ll_update_lsm_md(inode, md);
1951 #ifdef CONFIG_FS_POSIX_ACL
1952 if (body->mbo_valid & OBD_MD_FLACL) {
1953 spin_lock(&lli->lli_lock);
1954 if (lli->lli_posix_acl)
1955 posix_acl_release(lli->lli_posix_acl);
1956 lli->lli_posix_acl = md->posix_acl;
1957 spin_unlock(&lli->lli_lock);
1960 inode->i_ino = cl_fid_build_ino(&body->mbo_fid1,
1961 sbi->ll_flags & LL_SBI_32BIT_API);
1962 inode->i_generation = cl_fid_build_gen(&body->mbo_fid1);
1964 if (body->mbo_valid & OBD_MD_FLATIME) {
1965 if (body->mbo_atime > inode->i_atime.tv_sec)
1966 inode->i_atime.tv_sec = body->mbo_atime;
1967 lli->lli_atime = body->mbo_atime;
1970 if (body->mbo_valid & OBD_MD_FLMTIME) {
1971 if (body->mbo_mtime > inode->i_mtime.tv_sec) {
1973 "setting ino %lu mtime from %lld to %llu\n",
1974 inode->i_ino, (s64)inode->i_mtime.tv_sec,
1976 inode->i_mtime.tv_sec = body->mbo_mtime;
1978 lli->lli_mtime = body->mbo_mtime;
1981 if (body->mbo_valid & OBD_MD_FLCTIME) {
1982 if (body->mbo_ctime > inode->i_ctime.tv_sec)
1983 inode->i_ctime.tv_sec = body->mbo_ctime;
1984 lli->lli_ctime = body->mbo_ctime;
1987 /* Clear i_flags to remove S_NOSEC before permissions are updated */
1988 if (body->mbo_valid & OBD_MD_FLFLAGS)
1989 ll_update_inode_flags(inode, body->mbo_flags);
1990 if (body->mbo_valid & OBD_MD_FLMODE)
1991 inode->i_mode = (inode->i_mode & S_IFMT) |
1992 (body->mbo_mode & ~S_IFMT);
1994 if (body->mbo_valid & OBD_MD_FLTYPE)
1995 inode->i_mode = (inode->i_mode & ~S_IFMT) |
1996 (body->mbo_mode & S_IFMT);
1998 LASSERT(inode->i_mode != 0);
1999 if (S_ISREG(inode->i_mode))
2000 inode->i_blkbits = min(PTLRPC_MAX_BRW_BITS + 1,
2001 LL_MAX_BLKSIZE_BITS);
2003 inode->i_blkbits = inode->i_sb->s_blocksize_bits;
2005 if (body->mbo_valid & OBD_MD_FLUID)
2006 inode->i_uid = make_kuid(&init_user_ns, body->mbo_uid);
2007 if (body->mbo_valid & OBD_MD_FLGID)
2008 inode->i_gid = make_kgid(&init_user_ns, body->mbo_gid);
2009 if (body->mbo_valid & OBD_MD_FLPROJID)
2010 lli->lli_projid = body->mbo_projid;
2011 if (body->mbo_valid & OBD_MD_FLNLINK)
2012 set_nlink(inode, body->mbo_nlink);
2013 if (body->mbo_valid & OBD_MD_FLRDEV)
2014 inode->i_rdev = old_decode_dev(body->mbo_rdev);
2016 if (body->mbo_valid & OBD_MD_FLID) {
2017 /* FID shouldn't be changed! */
2018 if (fid_is_sane(&lli->lli_fid)) {
2019 LASSERTF(lu_fid_eq(&lli->lli_fid, &body->mbo_fid1),
2020 "Trying to change FID "DFID
2021 " to the "DFID", inode "DFID"(%p)\n",
2022 PFID(&lli->lli_fid), PFID(&body->mbo_fid1),
2023 PFID(ll_inode2fid(inode)), inode);
2025 lli->lli_fid = body->mbo_fid1;
2029 LASSERT(fid_seq(&lli->lli_fid) != 0);
2031 if (body->mbo_valid & OBD_MD_FLSIZE) {
2032 i_size_write(inode, body->mbo_size);
2034 CDEBUG(D_VFSTRACE, "inode="DFID", updating i_size %llu\n",
2035 PFID(ll_inode2fid(inode)),
2036 (unsigned long long)body->mbo_size);
2038 if (body->mbo_valid & OBD_MD_FLBLOCKS)
2039 inode->i_blocks = body->mbo_blocks;
2042 if (body->mbo_valid & OBD_MD_TSTATE) {
2043 /* Set LLIF_FILE_RESTORING if restore ongoing and
2044 * clear it when done to ensure to start again
2045 * glimpsing updated attrs
2047 if (body->mbo_t_state & MS_RESTORE)
2048 ll_file_set_flag(lli, LLIF_FILE_RESTORING);
2050 ll_file_clear_flag(lli, LLIF_FILE_RESTORING);
2056 void ll_truncate_inode_pages_final(struct inode *inode)
2058 struct address_space *mapping = &inode->i_data;
2059 unsigned long nrpages;
2060 unsigned long flags;
2062 truncate_inode_pages_final(mapping);
2064 /* Workaround for LU-118: Note nrpages may not be totally updated when
2065 * truncate_inode_pages() returns, as there can be a page in the process
2066 * of deletion (inside __delete_from_page_cache()) in the specified
2067 * range. Thus mapping->nrpages can be non-zero when this function
2068 * returns even after truncation of the whole mapping. Only do this if
2069 * npages isn't already zero.
2071 nrpages = mapping->nrpages;
2073 xa_lock_irqsave(&mapping->i_pages, flags);
2074 nrpages = mapping->nrpages;
2075 xa_unlock_irqrestore(&mapping->i_pages, flags);
2076 } /* Workaround end */
2078 LASSERTF(nrpages == 0, "%s: inode="DFID"(%p) nrpages=%lu, "
2079 "see https://jira.whamcloud.com/browse/LU-118\n",
2080 ll_get_fsname(inode->i_sb, NULL, 0),
2081 PFID(ll_inode2fid(inode)), inode, nrpages);
2084 int ll_read_inode2(struct inode *inode, void *opaque)
2086 struct lustre_md *md = opaque;
2087 struct ll_inode_info *lli = ll_i2info(inode);
2091 CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p)\n",
2092 PFID(&lli->lli_fid), inode);
2094 /* Core attributes from the MDS first. This is a new inode, and
2095 * the VFS doesn't zero times in the core inode so we have to do
2096 * it ourselves. They will be overwritten by either MDS or OST
2097 * attributes - we just need to make sure they aren't newer.
2099 inode->i_mtime.tv_sec = 0;
2100 inode->i_atime.tv_sec = 0;
2101 inode->i_ctime.tv_sec = 0;
2103 rc = ll_update_inode(inode, md);
2107 /* OIDEBUG(inode); */
2109 #ifdef HAVE_BACKING_DEV_INFO
2110 /* initializing backing dev info. */
2111 inode->i_mapping->backing_dev_info = &s2lsi(inode->i_sb)->lsi_bdi;
2113 if (S_ISREG(inode->i_mode)) {
2114 struct ll_sb_info *sbi = ll_i2sbi(inode);
2115 inode->i_op = &ll_file_inode_operations;
2116 inode->i_fop = sbi->ll_fop;
2117 inode->i_mapping->a_ops = (struct address_space_operations *)&ll_aops;
2119 } else if (S_ISDIR(inode->i_mode)) {
2120 inode->i_op = &ll_dir_inode_operations;
2121 inode->i_fop = &ll_dir_operations;
2123 } else if (S_ISLNK(inode->i_mode)) {
2124 inode->i_op = &ll_fast_symlink_inode_operations;
2127 inode->i_op = &ll_special_inode_operations;
2129 init_special_inode(inode, inode->i_mode,
2138 void ll_delete_inode(struct inode *inode)
2140 struct ll_inode_info *lli = ll_i2info(inode);
2143 if (S_ISREG(inode->i_mode) && lli->lli_clob != NULL) {
2144 /* It is last chance to write out dirty pages,
2145 * otherwise we may lose data while umount.
2147 * If i_nlink is 0 then just discard data. This is safe because
2148 * local inode gets i_nlink 0 from server only for the last
2149 * unlink, so that file is not opened somewhere else
2151 cl_sync_file_range(inode, 0, OBD_OBJECT_EOF, inode->i_nlink ?
2152 CL_FSYNC_LOCAL : CL_FSYNC_DISCARD, 1);
2155 ll_truncate_inode_pages_final(inode);
2157 #ifdef HAVE_SBOPS_EVICT_INODE
2158 ll_clear_inode(inode);
2165 int ll_iocontrol(struct inode *inode, struct file *file,
2166 unsigned int cmd, unsigned long arg)
2168 struct ll_sb_info *sbi = ll_i2sbi(inode);
2169 struct ptlrpc_request *req = NULL;
2174 case FS_IOC_GETFLAGS: {
2175 struct mdt_body *body;
2176 struct md_op_data *op_data;
2178 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL,
2179 0, 0, LUSTRE_OPC_ANY,
2181 if (IS_ERR(op_data))
2182 RETURN(PTR_ERR(op_data));
2184 op_data->op_valid = OBD_MD_FLFLAGS;
2185 rc = md_getattr(sbi->ll_md_exp, op_data, &req);
2186 ll_finish_md_op_data(op_data);
2188 CERROR("%s: failure inode "DFID": rc = %d\n",
2189 sbi->ll_md_exp->exp_obd->obd_name,
2190 PFID(ll_inode2fid(inode)), rc);
2194 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
2196 flags = body->mbo_flags;
2198 ptlrpc_req_finished(req);
2200 RETURN(put_user(flags, (int __user *)arg));
2202 case FS_IOC_SETFLAGS: {
2204 struct md_op_data *op_data;
2205 struct cl_object *obj;
2206 struct fsxattr fa = { 0 };
2208 if (get_user(flags, (int __user *)arg))
2211 fa.fsx_projid = ll_i2info(inode)->lli_projid;
2212 if (flags & LUSTRE_PROJINHERIT_FL)
2213 fa.fsx_xflags = FS_XFLAG_PROJINHERIT;
2215 rc = ll_ioctl_check_project(inode, &fa);
2219 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
2220 LUSTRE_OPC_ANY, NULL);
2221 if (IS_ERR(op_data))
2222 RETURN(PTR_ERR(op_data));
2224 op_data->op_attr_flags = flags;
2225 op_data->op_xvalid |= OP_XVALID_FLAGS;
2226 rc = md_setattr(sbi->ll_md_exp, op_data, NULL, 0, &req);
2227 ll_finish_md_op_data(op_data);
2228 ptlrpc_req_finished(req);
2232 ll_update_inode_flags(inode, flags);
2234 obj = ll_i2info(inode)->lli_clob;
2238 OBD_ALLOC_PTR(attr);
2242 rc = cl_setattr_ost(obj, attr, OP_XVALID_FLAGS, flags);
2254 int ll_flush_ctx(struct inode *inode)
2256 struct ll_sb_info *sbi = ll_i2sbi(inode);
2258 CDEBUG(D_SEC, "flush context for user %d\n",
2259 from_kuid(&init_user_ns, current_uid()));
2261 obd_set_info_async(NULL, sbi->ll_md_exp,
2262 sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
2264 obd_set_info_async(NULL, sbi->ll_dt_exp,
2265 sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
2270 /* umount -f client means force down, don't save state */
2271 void ll_umount_begin(struct super_block *sb)
2273 struct ll_sb_info *sbi = ll_s2sbi(sb);
2274 struct obd_device *obd;
2275 struct obd_ioctl_data *ioc_data;
2276 struct l_wait_info lwi;
2277 wait_queue_head_t waitq;
2280 CDEBUG(D_VFSTRACE, "VFS Op: superblock %p count %d active %d\n", sb,
2281 sb->s_count, atomic_read(&sb->s_active));
2283 obd = class_exp2obd(sbi->ll_md_exp);
2285 CERROR("Invalid MDC connection handle %#llx\n",
2286 sbi->ll_md_exp->exp_handle.h_cookie);
2292 obd = class_exp2obd(sbi->ll_dt_exp);
2294 CERROR("Invalid LOV connection handle %#llx\n",
2295 sbi->ll_dt_exp->exp_handle.h_cookie);
2301 OBD_ALLOC_PTR(ioc_data);
2303 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_md_exp,
2304 sizeof *ioc_data, ioc_data, NULL);
2306 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_dt_exp,
2307 sizeof *ioc_data, ioc_data, NULL);
2309 OBD_FREE_PTR(ioc_data);
2312 /* Really, we'd like to wait until there are no requests outstanding,
2313 * and then continue. For now, we just periodically checking for vfs
2314 * to decrement mnt_cnt and hope to finish it within 10sec.
2316 init_waitqueue_head(&waitq);
2317 lwi = LWI_TIMEOUT_INTERVAL(cfs_time_seconds(10),
2318 cfs_time_seconds(1), NULL, NULL);
2319 l_wait_event(waitq, may_umount(sbi->ll_mnt.mnt), &lwi);
2324 int ll_remount_fs(struct super_block *sb, int *flags, char *data)
2326 struct ll_sb_info *sbi = ll_s2sbi(sb);
2327 char *profilenm = get_profile_name(sb);
2331 if ((*flags & MS_RDONLY) != (sb->s_flags & SB_RDONLY)) {
2332 read_only = *flags & MS_RDONLY;
2333 err = obd_set_info_async(NULL, sbi->ll_md_exp,
2334 sizeof(KEY_READ_ONLY),
2335 KEY_READ_ONLY, sizeof(read_only),
2338 LCONSOLE_WARN("Failed to remount %s %s (%d)\n",
2339 profilenm, read_only ?
2340 "read-only" : "read-write", err);
2345 sb->s_flags |= SB_RDONLY;
2347 sb->s_flags &= ~SB_RDONLY;
2349 if (sbi->ll_flags & LL_SBI_VERBOSE)
2350 LCONSOLE_WARN("Remounted %s %s\n", profilenm,
2351 read_only ? "read-only" : "read-write");
2357 * Cleanup the open handle that is cached on MDT-side.
2359 * For open case, the client side open handling thread may hit error
2360 * after the MDT grant the open. Under such case, the client should
2361 * send close RPC to the MDT as cleanup; otherwise, the open handle
2362 * on the MDT will be leaked there until the client umount or evicted.
2364 * In further, if someone unlinked the file, because the open handle
2365 * holds the reference on such file/object, then it will block the
2366 * subsequent threads that want to locate such object via FID.
2368 * \param[in] sb super block for this file-system
2369 * \param[in] open_req pointer to the original open request
2371 void ll_open_cleanup(struct super_block *sb, struct ptlrpc_request *open_req)
2373 struct mdt_body *body;
2374 struct md_op_data *op_data;
2375 struct ptlrpc_request *close_req = NULL;
2376 struct obd_export *exp = ll_s2sbi(sb)->ll_md_exp;
2379 body = req_capsule_server_get(&open_req->rq_pill, &RMF_MDT_BODY);
2380 OBD_ALLOC_PTR(op_data);
2381 if (op_data == NULL) {
2382 CWARN("%s: cannot allocate op_data to release open handle for "
2384 ll_get_fsname(sb, NULL, 0), PFID(&body->mbo_fid1));
2389 op_data->op_fid1 = body->mbo_fid1;
2390 op_data->op_open_handle = body->mbo_open_handle;
2391 op_data->op_mod_time = ktime_get_real_seconds();
2392 md_close(exp, op_data, NULL, &close_req);
2393 ptlrpc_req_finished(close_req);
2394 ll_finish_md_op_data(op_data);
2399 int ll_prep_inode(struct inode **inode, struct ptlrpc_request *req,
2400 struct super_block *sb, struct lookup_intent *it)
2402 struct ll_sb_info *sbi = NULL;
2403 struct lustre_md md = { NULL };
2407 LASSERT(*inode || sb);
2408 sbi = sb ? ll_s2sbi(sb) : ll_i2sbi(*inode);
2409 rc = md_get_lustre_md(sbi->ll_md_exp, req, sbi->ll_dt_exp,
2410 sbi->ll_md_exp, &md);
2415 rc = ll_update_inode(*inode, &md);
2419 LASSERT(sb != NULL);
2422 * At this point server returns to client's same fid as client
2423 * generated for creating. So using ->fid1 is okay here.
2425 if (!fid_is_sane(&md.body->mbo_fid1)) {
2426 CERROR("%s: Fid is insane "DFID"\n",
2427 ll_get_fsname(sb, NULL, 0),
2428 PFID(&md.body->mbo_fid1));
2429 GOTO(out, rc = -EINVAL);
2432 *inode = ll_iget(sb, cl_fid_build_ino(&md.body->mbo_fid1,
2433 sbi->ll_flags & LL_SBI_32BIT_API),
2435 if (IS_ERR(*inode)) {
2436 #ifdef CONFIG_FS_POSIX_ACL
2438 posix_acl_release(md.posix_acl);
2439 md.posix_acl = NULL;
2442 rc = IS_ERR(*inode) ? PTR_ERR(*inode) : -ENOMEM;
2444 CERROR("new_inode -fatal: rc %d\n", rc);
2449 /* Handling piggyback layout lock.
2450 * Layout lock can be piggybacked by getattr and open request.
2451 * The lsm can be applied to inode only if it comes with a layout lock
2452 * otherwise correct layout may be overwritten, for example:
2453 * 1. proc1: mdt returns a lsm but not granting layout
2454 * 2. layout was changed by another client
2455 * 3. proc2: refresh layout and layout lock granted
2456 * 4. proc1: to apply a stale layout */
2457 if (it != NULL && it->it_lock_mode != 0) {
2458 struct lustre_handle lockh;
2459 struct ldlm_lock *lock;
2461 lockh.cookie = it->it_lock_handle;
2462 lock = ldlm_handle2lock(&lockh);
2463 LASSERT(lock != NULL);
2464 if (ldlm_has_layout(lock)) {
2465 struct cl_object_conf conf;
2467 memset(&conf, 0, sizeof(conf));
2468 conf.coc_opc = OBJECT_CONF_SET;
2469 conf.coc_inode = *inode;
2470 conf.coc_lock = lock;
2471 conf.u.coc_layout = md.layout;
2472 (void)ll_layout_conf(*inode, &conf);
2474 LDLM_LOCK_PUT(lock);
2480 md_free_lustre_md(sbi->ll_md_exp, &md);
2483 if (rc != 0 && it != NULL && it->it_op & IT_OPEN) {
2484 ll_intent_drop_lock(it);
2485 ll_open_cleanup(sb != NULL ? sb : (*inode)->i_sb, req);
2491 int ll_obd_statfs(struct inode *inode, void __user *arg)
2493 struct ll_sb_info *sbi = NULL;
2494 struct obd_export *exp;
2496 struct obd_ioctl_data *data = NULL;
2500 if (!inode || !(sbi = ll_i2sbi(inode)))
2501 GOTO(out_statfs, rc = -EINVAL);
2503 rc = obd_ioctl_getdata(&buf, &len, arg);
2505 GOTO(out_statfs, rc);
2508 if (!data->ioc_inlbuf1 || !data->ioc_inlbuf2 ||
2509 !data->ioc_pbuf1 || !data->ioc_pbuf2)
2510 GOTO(out_statfs, rc = -EINVAL);
2512 if (data->ioc_inllen1 != sizeof(__u32) ||
2513 data->ioc_inllen2 != sizeof(__u32) ||
2514 data->ioc_plen1 != sizeof(struct obd_statfs) ||
2515 data->ioc_plen2 != sizeof(struct obd_uuid))
2516 GOTO(out_statfs, rc = -EINVAL);
2518 memcpy(&type, data->ioc_inlbuf1, sizeof(__u32));
2519 if (type & LL_STATFS_LMV)
2520 exp = sbi->ll_md_exp;
2521 else if (type & LL_STATFS_LOV)
2522 exp = sbi->ll_dt_exp;
2524 GOTO(out_statfs, rc = -ENODEV);
2526 rc = obd_iocontrol(IOC_OBD_STATFS, exp, len, buf, NULL);
2528 GOTO(out_statfs, rc);
2530 OBD_FREE_LARGE(buf, len);
2535 * this is normally called in ll_fini_md_op_data(), but sometimes it needs to
2536 * be called early to avoid deadlock.
2538 void ll_unlock_md_op_lsm(struct md_op_data *op_data)
2540 if (op_data->op_mea2_sem) {
2541 up_read(op_data->op_mea2_sem);
2542 op_data->op_mea2_sem = NULL;
2545 if (op_data->op_mea1_sem) {
2546 up_read(op_data->op_mea1_sem);
2547 op_data->op_mea1_sem = NULL;
2551 /* this function prepares md_op_data hint for passing it down to MD stack. */
2552 struct md_op_data *ll_prep_md_op_data(struct md_op_data *op_data,
2553 struct inode *i1, struct inode *i2,
2554 const char *name, size_t namelen,
2555 __u32 mode, __u32 opc, void *data)
2557 LASSERT(i1 != NULL);
2560 /* Do not reuse namelen for something else. */
2562 return ERR_PTR(-EINVAL);
2564 if (namelen > ll_i2sbi(i1)->ll_namelen)
2565 return ERR_PTR(-ENAMETOOLONG);
2567 /* "/" is not valid name, but it's allowed */
2568 if (!lu_name_is_valid_2(name, namelen) &&
2569 strncmp("/", name, namelen) != 0)
2570 return ERR_PTR(-EINVAL);
2573 if (op_data == NULL)
2574 OBD_ALLOC_PTR(op_data);
2576 if (op_data == NULL)
2577 return ERR_PTR(-ENOMEM);
2579 ll_i2gids(op_data->op_suppgids, i1, i2);
2580 op_data->op_fid1 = *ll_inode2fid(i1);
2581 op_data->op_default_stripe_offset = -1;
2583 if (S_ISDIR(i1->i_mode)) {
2584 down_read(&ll_i2info(i1)->lli_lsm_sem);
2585 op_data->op_mea1_sem = &ll_i2info(i1)->lli_lsm_sem;
2586 op_data->op_mea1 = ll_i2info(i1)->lli_lsm_md;
2587 if (opc == LUSTRE_OPC_MKDIR)
2588 op_data->op_default_stripe_offset =
2589 ll_i2info(i1)->lli_def_stripe_offset;
2593 op_data->op_fid2 = *ll_inode2fid(i2);
2594 if (S_ISDIR(i2->i_mode)) {
2596 down_read(&ll_i2info(i2)->lli_lsm_sem);
2597 op_data->op_mea2_sem =
2598 &ll_i2info(i2)->lli_lsm_sem;
2600 op_data->op_mea2 = ll_i2info(i2)->lli_lsm_md;
2603 fid_zero(&op_data->op_fid2);
2606 if (ll_i2sbi(i1)->ll_flags & LL_SBI_64BIT_HASH)
2607 op_data->op_cli_flags |= CLI_HASH64;
2609 if (ll_need_32bit_api(ll_i2sbi(i1)))
2610 op_data->op_cli_flags |= CLI_API32;
2612 op_data->op_name = name;
2613 op_data->op_namelen = namelen;
2614 op_data->op_mode = mode;
2615 op_data->op_mod_time = ktime_get_real_seconds();
2616 op_data->op_fsuid = from_kuid(&init_user_ns, current_fsuid());
2617 op_data->op_fsgid = from_kgid(&init_user_ns, current_fsgid());
2618 op_data->op_cap = cfs_curproc_cap_pack();
2619 op_data->op_mds = 0;
2620 if ((opc == LUSTRE_OPC_CREATE) && (name != NULL) &&
2621 filename_is_volatile(name, namelen, &op_data->op_mds)) {
2622 op_data->op_bias |= MDS_CREATE_VOLATILE;
2624 op_data->op_data = data;
2629 void ll_finish_md_op_data(struct md_op_data *op_data)
2631 ll_unlock_md_op_lsm(op_data);
2632 security_release_secctx(op_data->op_file_secctx,
2633 op_data->op_file_secctx_size);
2634 OBD_FREE_PTR(op_data);
2637 #ifdef HAVE_SUPEROPS_USE_DENTRY
2638 int ll_show_options(struct seq_file *seq, struct dentry *dentry)
2640 int ll_show_options(struct seq_file *seq, struct vfsmount *vfs)
2643 struct ll_sb_info *sbi;
2645 #ifdef HAVE_SUPEROPS_USE_DENTRY
2646 LASSERT((seq != NULL) && (dentry != NULL));
2647 sbi = ll_s2sbi(dentry->d_sb);
2649 LASSERT((seq != NULL) && (vfs != NULL));
2650 sbi = ll_s2sbi(vfs->mnt_sb);
2653 if (sbi->ll_flags & LL_SBI_NOLCK)
2654 seq_puts(seq, ",nolock");
2656 /* "flock" is the default since 2.13, but it wasn't for many years,
2657 * so it is still useful to print this to show it is enabled.
2658 * Start to print "noflock" so it is now clear when flock is disabled.
2660 if (sbi->ll_flags & LL_SBI_FLOCK)
2661 seq_puts(seq, ",flock");
2662 else if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
2663 seq_puts(seq, ",localflock");
2665 seq_puts(seq, ",noflock");
2667 if (sbi->ll_flags & LL_SBI_USER_XATTR)
2668 seq_puts(seq, ",user_xattr");
2670 if (sbi->ll_flags & LL_SBI_LAZYSTATFS)
2671 seq_puts(seq, ",lazystatfs");
2673 if (sbi->ll_flags & LL_SBI_USER_FID2PATH)
2674 seq_puts(seq, ",user_fid2path");
2676 if (sbi->ll_flags & LL_SBI_ALWAYS_PING)
2677 seq_puts(seq, ",always_ping");
2683 * Get obd name by cmd, and copy out to user space
2685 int ll_get_obd_name(struct inode *inode, unsigned int cmd, unsigned long arg)
2687 struct ll_sb_info *sbi = ll_i2sbi(inode);
2688 struct obd_device *obd;
2691 if (cmd == OBD_IOC_GETDTNAME)
2692 obd = class_exp2obd(sbi->ll_dt_exp);
2693 else if (cmd == OBD_IOC_GETMDNAME)
2694 obd = class_exp2obd(sbi->ll_md_exp);
2701 if (copy_to_user((void __user *)arg, obd->obd_name,
2702 strlen(obd->obd_name) + 1))
2709 * Get lustre file system name by \a sbi. If \a buf is provided(non-NULL), the
2710 * fsname will be returned in this buffer; otherwise, a static buffer will be
2711 * used to store the fsname and returned to caller.
2713 char *ll_get_fsname(struct super_block *sb, char *buf, int buflen)
2715 static char fsname_static[MTI_NAME_MAXLEN];
2716 struct lustre_sb_info *lsi = s2lsi(sb);
2721 /* this means the caller wants to use static buffer
2722 * and it doesn't care about race. Usually this is
2723 * in error reporting path */
2724 buf = fsname_static;
2725 buflen = sizeof(fsname_static);
2728 len = strlen(lsi->lsi_lmd->lmd_profile);
2729 ptr = strrchr(lsi->lsi_lmd->lmd_profile, '-');
2730 if (ptr && (strcmp(ptr, "-client") == 0))
2733 if (unlikely(len >= buflen))
2735 strncpy(buf, lsi->lsi_lmd->lmd_profile, len);
2741 static char* ll_d_path(struct dentry *dentry, char *buf, int bufsize)
2748 p.mnt = current->fs->root.mnt;
2750 path = d_path(&p, buf, bufsize);
2755 void ll_dirty_page_discard_warn(struct page *page, int ioret)
2757 char *buf, *path = NULL;
2758 struct dentry *dentry = NULL;
2759 struct inode *inode = page->mapping->host;
2761 /* this can be called inside spin lock so use GFP_ATOMIC. */
2762 buf = (char *)__get_free_page(GFP_ATOMIC);
2764 dentry = d_find_alias(page->mapping->host);
2766 path = ll_d_path(dentry, buf, PAGE_SIZE);
2770 "%s: dirty page discard: %s/fid: "DFID"/%s may get corrupted "
2771 "(rc %d)\n", ll_get_fsname(page->mapping->host->i_sb, NULL, 0),
2772 s2lsi(page->mapping->host->i_sb)->lsi_lmd->lmd_dev,
2773 PFID(ll_inode2fid(inode)),
2774 (path && !IS_ERR(path)) ? path : "", ioret);
2780 free_page((unsigned long)buf);
2783 ssize_t ll_copy_user_md(const struct lov_user_md __user *md,
2784 struct lov_user_md **kbuf)
2786 struct lov_user_md lum;
2790 if (copy_from_user(&lum, md, sizeof(lum)))
2793 lum_size = ll_lov_user_md_size(&lum);
2797 OBD_ALLOC_LARGE(*kbuf, lum_size);
2801 if (copy_from_user(*kbuf, md, lum_size) != 0) {
2802 OBD_FREE_LARGE(*kbuf, lum_size);
2810 * Compute llite root squash state after a change of root squash
2811 * configuration setting or add/remove of a lnet nid
2813 void ll_compute_rootsquash_state(struct ll_sb_info *sbi)
2815 struct root_squash_info *squash = &sbi->ll_squash;
2818 struct lnet_process_id id;
2820 /* Update norootsquash flag */
2821 down_write(&squash->rsi_sem);
2822 if (list_empty(&squash->rsi_nosquash_nids))
2823 sbi->ll_flags &= ~LL_SBI_NOROOTSQUASH;
2825 /* Do not apply root squash as soon as one of our NIDs is
2826 * in the nosquash_nids list */
2829 while (LNetGetId(i++, &id) != -ENOENT) {
2830 if (id.nid == LNET_NID_LO_0)
2832 if (cfs_match_nid(id.nid, &squash->rsi_nosquash_nids)) {
2838 sbi->ll_flags |= LL_SBI_NOROOTSQUASH;
2840 sbi->ll_flags &= ~LL_SBI_NOROOTSQUASH;
2842 up_write(&squash->rsi_sem);
2846 * Parse linkea content to extract information about a given hardlink
2848 * \param[in] ldata - Initialized linkea data
2849 * \param[in] linkno - Link identifier
2850 * \param[out] parent_fid - The entry's parent FID
2851 * \param[out] ln - Entry name destination buffer
2853 * \retval 0 on success
2854 * \retval Appropriate negative error code on failure
2856 static int ll_linkea_decode(struct linkea_data *ldata, unsigned int linkno,
2857 struct lu_fid *parent_fid, struct lu_name *ln)
2863 rc = linkea_init_with_rec(ldata);
2867 if (linkno >= ldata->ld_leh->leh_reccount)
2868 /* beyond last link */
2871 linkea_first_entry(ldata);
2872 for (idx = 0; ldata->ld_lee != NULL; idx++) {
2873 linkea_entry_unpack(ldata->ld_lee, &ldata->ld_reclen, ln,
2878 linkea_next_entry(ldata);
2888 * Get parent FID and name of an identified link. Operation is performed for
2889 * a given link number, letting the caller iterate over linkno to list one or
2890 * all links of an entry.
2892 * \param[in] file - File descriptor against which to perform the operation
2893 * \param[in,out] arg - User-filled structure containing the linkno to operate
2894 * on and the available size. It is eventually filled with
2895 * the requested information or left untouched on error
2897 * \retval - 0 on success
2898 * \retval - Appropriate negative error code on failure
2900 int ll_getparent(struct file *file, struct getparent __user *arg)
2902 struct inode *inode = file_inode(file);
2903 struct linkea_data *ldata;
2904 struct lu_buf buf = LU_BUF_NULL;
2906 struct lu_fid parent_fid;
2913 if (!cfs_capable(CFS_CAP_DAC_READ_SEARCH) &&
2914 !(ll_i2sbi(inode)->ll_flags & LL_SBI_USER_FID2PATH))
2917 if (get_user(name_size, &arg->gp_name_size))
2920 if (get_user(linkno, &arg->gp_linkno))
2923 if (name_size > PATH_MAX)
2926 OBD_ALLOC(ldata, sizeof(*ldata));
2930 rc = linkea_data_new(ldata, &buf);
2932 GOTO(ldata_free, rc);
2934 #ifdef HAVE_XATTR_HANDLER_FLAGS
2935 rc = ll_xattr_list(inode, XATTR_NAME_LINK, XATTR_TRUSTED_T, buf.lb_buf,
2936 buf.lb_len, OBD_MD_FLXATTR);
2938 rc = ll_getxattr(file_dentry(file), XATTR_NAME_LINK, buf.lb_buf,
2940 #endif /* HAVE_XATTR_HANDLER_FLAGS */
2944 rc = ll_linkea_decode(ldata, linkno, &parent_fid, &ln);
2948 if (ln.ln_namelen >= name_size)
2949 GOTO(lb_free, rc = -EOVERFLOW);
2951 if (copy_to_user(&arg->gp_fid, &parent_fid, sizeof(arg->gp_fid)))
2952 GOTO(lb_free, rc = -EFAULT);
2954 if (copy_to_user(&arg->gp_name, ln.ln_name, ln.ln_namelen))
2955 GOTO(lb_free, rc = -EFAULT);
2957 if (put_user('\0', arg->gp_name + ln.ln_namelen))
2958 GOTO(lb_free, rc = -EFAULT);
2963 OBD_FREE(ldata, sizeof(*ldata));