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 #include <lustre_ha.h>
55 #include <lustre_dlm.h>
56 #include <lprocfs_status.h>
57 #include <lustre_disk.h>
58 #include <uapi/linux/lustre/lustre_param.h>
59 #include <lustre_log.h>
60 #include <cl_object.h>
61 #include <obd_cksum.h>
62 #include "llite_internal.h"
64 struct kmem_cache *ll_file_data_slab;
67 #define log2(n) ffz(~(n))
70 static struct ll_sb_info *ll_init_sbi(void)
72 struct ll_sb_info *sbi = NULL;
74 unsigned long lru_page_max;
83 spin_lock_init(&sbi->ll_lock);
84 mutex_init(&sbi->ll_lco.lco_lock);
85 spin_lock_init(&sbi->ll_pp_extent_lock);
86 spin_lock_init(&sbi->ll_process_lock);
87 sbi->ll_rw_stats_on = 0;
90 pages = si.totalram - si.totalhigh;
91 lru_page_max = pages / 2;
93 /* initialize ll_cache data */
94 sbi->ll_cache = cl_cache_init(lru_page_max);
95 if (sbi->ll_cache == NULL) {
96 OBD_FREE(sbi, sizeof(*sbi));
100 sbi->ll_ra_info.ra_max_pages_per_file = min(pages / 32,
101 SBI_DEFAULT_READAHEAD_MAX);
102 sbi->ll_ra_info.ra_max_pages = sbi->ll_ra_info.ra_max_pages_per_file;
103 sbi->ll_ra_info.ra_max_read_ahead_whole_pages = -1;
105 sbi->ll_flags |= LL_SBI_VERBOSE;
106 #ifdef ENABLE_CHECKSUM
107 sbi->ll_flags |= LL_SBI_CHECKSUM;
110 sbi->ll_flags |= LL_SBI_FLOCK;
113 #ifdef HAVE_LRU_RESIZE_SUPPORT
114 sbi->ll_flags |= LL_SBI_LRU_RESIZE;
116 sbi->ll_flags |= LL_SBI_LAZYSTATFS;
118 for (i = 0; i <= LL_PROCESS_HIST_MAX; i++) {
119 spin_lock_init(&sbi->ll_rw_extents_info.pp_extents[i].
121 spin_lock_init(&sbi->ll_rw_extents_info.pp_extents[i].
125 /* metadata statahead is enabled by default */
126 sbi->ll_sa_running_max = LL_SA_RUNNING_DEF;
127 sbi->ll_sa_max = LL_SA_RPC_DEF;
128 atomic_set(&sbi->ll_sa_total, 0);
129 atomic_set(&sbi->ll_sa_wrong, 0);
130 atomic_set(&sbi->ll_sa_running, 0);
131 atomic_set(&sbi->ll_agl_total, 0);
132 sbi->ll_flags |= LL_SBI_AGL_ENABLED;
133 sbi->ll_flags |= LL_SBI_FAST_READ;
134 sbi->ll_flags |= LL_SBI_TINY_WRITE;
137 sbi->ll_squash.rsi_uid = 0;
138 sbi->ll_squash.rsi_gid = 0;
139 INIT_LIST_HEAD(&sbi->ll_squash.rsi_nosquash_nids);
140 init_rwsem(&sbi->ll_squash.rsi_sem);
142 /* Per-filesystem file heat */
143 sbi->ll_heat_decay_weight = SBI_DEFAULT_HEAT_DECAY_WEIGHT;
144 sbi->ll_heat_period_second = SBI_DEFAULT_HEAT_PERIOD_SECOND;
148 static void ll_free_sbi(struct super_block *sb)
150 struct ll_sb_info *sbi = ll_s2sbi(sb);
154 if (!list_empty(&sbi->ll_squash.rsi_nosquash_nids))
155 cfs_free_nidlist(&sbi->ll_squash.rsi_nosquash_nids);
156 if (sbi->ll_cache != NULL) {
157 cl_cache_decref(sbi->ll_cache);
158 sbi->ll_cache = NULL;
160 OBD_FREE(sbi, sizeof(*sbi));
165 static int client_common_fill_super(struct super_block *sb, char *md, char *dt,
166 struct vfsmount *mnt)
168 struct inode *root = NULL;
169 struct ll_sb_info *sbi = ll_s2sbi(sb);
170 struct obd_statfs *osfs = NULL;
171 struct ptlrpc_request *request = NULL;
172 struct obd_connect_data *data = NULL;
173 struct obd_uuid *uuid;
174 struct md_op_data *op_data;
175 struct lustre_md lmd;
177 int size, err, checksum;
180 sbi->ll_md_obd = class_name2obd(md);
181 if (!sbi->ll_md_obd) {
182 CERROR("MD %s: not setup or attached\n", md);
196 /* pass client page size via ocd_grant_blkbits, the server should report
197 * back its backend blocksize for grant calculation purpose */
198 data->ocd_grant_blkbits = PAGE_SHIFT;
200 /* indicate MDT features supported by this client */
201 data->ocd_connect_flags = OBD_CONNECT_IBITS | OBD_CONNECT_NODEVOH |
202 OBD_CONNECT_ATTRFID | OBD_CONNECT_GRANT |
203 OBD_CONNECT_VERSION | OBD_CONNECT_BRW_SIZE |
204 OBD_CONNECT_SRVLOCK | OBD_CONNECT_TRUNCLOCK|
205 OBD_CONNECT_MDS_CAPA | OBD_CONNECT_OSS_CAPA |
206 OBD_CONNECT_CANCELSET | OBD_CONNECT_FID |
207 OBD_CONNECT_AT | OBD_CONNECT_LOV_V3 |
208 OBD_CONNECT_VBR | OBD_CONNECT_FULL20 |
209 OBD_CONNECT_64BITHASH |
210 OBD_CONNECT_EINPROGRESS |
211 OBD_CONNECT_JOBSTATS | OBD_CONNECT_LVB_TYPE |
212 OBD_CONNECT_LAYOUTLOCK | OBD_CONNECT_PINGLESS|
213 OBD_CONNECT_MAX_EASIZE |
214 OBD_CONNECT_FLOCK_DEAD |
215 OBD_CONNECT_DISP_STRIPE | OBD_CONNECT_LFSCK |
216 OBD_CONNECT_OPEN_BY_FID |
217 OBD_CONNECT_DIR_STRIPE |
218 OBD_CONNECT_BULK_MBITS | OBD_CONNECT_CKSUM |
219 OBD_CONNECT_SUBTREE |
220 OBD_CONNECT_MULTIMODRPCS |
221 OBD_CONNECT_GRANT_PARAM |
222 OBD_CONNECT_SHORTIO | OBD_CONNECT_FLAGS2;
224 data->ocd_connect_flags2 = OBD_CONNECT2_DIR_MIGRATE |
225 OBD_CONNECT2_SUM_STATFS |
227 OBD_CONNECT2_LOCK_CONVERT |
228 OBD_CONNECT2_ARCHIVE_ID_ARRAY |
230 OBD_CONNECT2_ASYNC_DISCARD;
232 #ifdef HAVE_LRU_RESIZE_SUPPORT
233 if (sbi->ll_flags & LL_SBI_LRU_RESIZE)
234 data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
236 #ifdef CONFIG_FS_POSIX_ACL
237 data->ocd_connect_flags |= OBD_CONNECT_ACL | OBD_CONNECT_UMASK |
238 OBD_CONNECT_LARGE_ACL;
241 data->ocd_cksum_types = obd_cksum_types_supported_client();
243 if (OBD_FAIL_CHECK(OBD_FAIL_MDC_LIGHTWEIGHT))
244 /* flag mdc connection as lightweight, only used for test
245 * purpose, use with care */
246 data->ocd_connect_flags |= OBD_CONNECT_LIGHTWEIGHT;
248 data->ocd_ibits_known = MDS_INODELOCK_FULL;
249 data->ocd_version = LUSTRE_VERSION_CODE;
251 if (sb->s_flags & MS_RDONLY)
252 data->ocd_connect_flags |= OBD_CONNECT_RDONLY;
253 if (sbi->ll_flags & LL_SBI_USER_XATTR)
254 data->ocd_connect_flags |= OBD_CONNECT_XATTR;
257 /* Setting this indicates we correctly support S_NOSEC (See kernel
258 * commit 9e1f1de02c2275d7172e18dc4e7c2065777611bf)
260 sb->s_flags |= MS_NOSEC;
263 if (sbi->ll_flags & LL_SBI_FLOCK)
264 sbi->ll_fop = &ll_file_operations_flock;
265 else if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
266 sbi->ll_fop = &ll_file_operations;
268 sbi->ll_fop = &ll_file_operations_noflock;
270 /* always ping even if server suppress_pings */
271 if (sbi->ll_flags & LL_SBI_ALWAYS_PING)
272 data->ocd_connect_flags &= ~OBD_CONNECT_PINGLESS;
274 obd_connect_set_secctx(data);
276 #if defined(CONFIG_SECURITY)
277 data->ocd_connect_flags2 |= OBD_CONNECT2_SELINUX_POLICY;
280 data->ocd_brw_size = MD_MAX_BRW_SIZE;
282 err = obd_connect(NULL, &sbi->ll_md_exp, sbi->ll_md_obd,
283 &sbi->ll_sb_uuid, data, sbi->ll_cache);
285 LCONSOLE_ERROR_MSG(0x14f, "An MDT (md %s) is performing "
286 "recovery, of which this client is not a "
287 "part. Please wait for recovery to complete,"
288 " abort, or time out.\n", md);
291 CERROR("cannot connect to %s: rc = %d\n", md, err);
295 sbi->ll_md_exp->exp_connect_data = *data;
297 err = obd_fid_init(sbi->ll_md_exp->exp_obd, sbi->ll_md_exp,
298 LUSTRE_SEQ_METADATA);
300 CERROR("%s: Can't init metadata layer FID infrastructure, "
301 "rc = %d\n", sbi->ll_md_exp->exp_obd->obd_name, err);
305 /* For mount, we only need fs info from MDT0, and also in DNE, it
306 * can make sure the client can be mounted as long as MDT0 is
308 err = obd_statfs(NULL, sbi->ll_md_exp, osfs,
309 ktime_get_seconds() - OBD_STATFS_CACHE_SECONDS,
310 OBD_STATFS_FOR_MDT0);
312 GOTO(out_md_fid, err);
314 /* This needs to be after statfs to ensure connect has finished.
315 * Note that "data" does NOT contain the valid connect reply.
316 * If connecting to a 1.8 server there will be no LMV device, so
317 * we can access the MDC export directly and exp_connect_flags will
318 * be non-zero, but if accessing an upgraded 2.1 server it will
319 * have the correct flags filled in.
320 * XXX: fill in the LMV exp_connect_flags from MDC(s). */
321 valid = exp_connect_flags(sbi->ll_md_exp) & CLIENT_CONNECT_MDT_REQD;
322 if (exp_connect_flags(sbi->ll_md_exp) != 0 &&
323 valid != CLIENT_CONNECT_MDT_REQD) {
326 OBD_ALLOC_WAIT(buf, PAGE_SIZE);
327 obd_connect_flags2str(buf, PAGE_SIZE,
328 valid ^ CLIENT_CONNECT_MDT_REQD, 0, ",");
329 LCONSOLE_ERROR_MSG(0x170, "Server %s does not support "
330 "feature(s) needed for correct operation "
331 "of this client (%s). Please upgrade "
332 "server or downgrade client.\n",
333 sbi->ll_md_exp->exp_obd->obd_name, buf);
334 OBD_FREE(buf, PAGE_SIZE);
335 GOTO(out_md_fid, err = -EPROTO);
338 size = sizeof(*data);
339 err = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_CONN_DATA),
340 KEY_CONN_DATA, &size, data);
342 CERROR("%s: Get connect data failed: rc = %d\n",
343 sbi->ll_md_exp->exp_obd->obd_name, err);
344 GOTO(out_md_fid, err);
347 LASSERT(osfs->os_bsize);
348 sb->s_blocksize = osfs->os_bsize;
349 sb->s_blocksize_bits = log2(osfs->os_bsize);
350 sb->s_magic = LL_SUPER_MAGIC;
351 sb->s_maxbytes = MAX_LFS_FILESIZE;
352 sbi->ll_namelen = osfs->os_namelen;
353 sbi->ll_mnt.mnt = current->fs->root.mnt;
355 if ((sbi->ll_flags & LL_SBI_USER_XATTR) &&
356 !(data->ocd_connect_flags & OBD_CONNECT_XATTR)) {
357 LCONSOLE_INFO("Disabling user_xattr feature because "
358 "it is not supported on the server\n");
359 sbi->ll_flags &= ~LL_SBI_USER_XATTR;
362 if (data->ocd_connect_flags & OBD_CONNECT_ACL) {
364 sb->s_flags |= MS_POSIXACL;
366 sbi->ll_flags |= LL_SBI_ACL;
368 LCONSOLE_INFO("client wants to enable acl, but mdt not!\n");
370 sb->s_flags &= ~MS_POSIXACL;
372 sbi->ll_flags &= ~LL_SBI_ACL;
375 if (data->ocd_connect_flags & OBD_CONNECT_64BITHASH)
376 sbi->ll_flags |= LL_SBI_64BIT_HASH;
378 if (data->ocd_connect_flags & OBD_CONNECT_LAYOUTLOCK)
379 sbi->ll_flags |= LL_SBI_LAYOUT_LOCK;
381 if (obd_connect_has_secctx(data))
382 sbi->ll_flags |= LL_SBI_FILE_SECCTX;
384 if (data->ocd_ibits_known & MDS_INODELOCK_XATTR) {
385 if (!(data->ocd_connect_flags & OBD_CONNECT_MAX_EASIZE)) {
386 LCONSOLE_INFO("%s: disabling xattr cache due to "
387 "unknown maximum xattr size.\n", dt);
388 } else if (!sbi->ll_xattr_cache_set) {
389 /* If xattr_cache is already set (no matter 0 or 1)
390 * during processing llog, it won't be enabled here. */
391 sbi->ll_flags |= LL_SBI_XATTR_CACHE;
392 sbi->ll_xattr_cache_enabled = 1;
396 sbi->ll_dt_obd = class_name2obd(dt);
397 if (!sbi->ll_dt_obd) {
398 CERROR("DT %s: not setup or attached\n", dt);
399 GOTO(out_md_fid, err = -ENODEV);
402 /* pass client page size via ocd_grant_blkbits, the server should report
403 * back its backend blocksize for grant calculation purpose */
404 data->ocd_grant_blkbits = PAGE_SHIFT;
406 /* indicate OST features supported by this client */
407 data->ocd_connect_flags = OBD_CONNECT_GRANT | OBD_CONNECT_VERSION |
408 OBD_CONNECT_REQPORTAL | OBD_CONNECT_BRW_SIZE |
409 OBD_CONNECT_CANCELSET | OBD_CONNECT_FID |
410 OBD_CONNECT_SRVLOCK | OBD_CONNECT_TRUNCLOCK|
411 OBD_CONNECT_AT | OBD_CONNECT_OSS_CAPA |
412 OBD_CONNECT_VBR | OBD_CONNECT_FULL20 |
413 OBD_CONNECT_64BITHASH | OBD_CONNECT_MAXBYTES |
414 OBD_CONNECT_EINPROGRESS |
415 OBD_CONNECT_JOBSTATS | OBD_CONNECT_LVB_TYPE |
416 OBD_CONNECT_LAYOUTLOCK |
417 OBD_CONNECT_PINGLESS | OBD_CONNECT_LFSCK |
418 OBD_CONNECT_BULK_MBITS | OBD_CONNECT_SHORTIO |
419 OBD_CONNECT_FLAGS2 | OBD_CONNECT_GRANT_SHRINK;
421 /* The client currently advertises support for OBD_CONNECT_LOCKAHEAD_OLD so it
422 * can interoperate with an older version of lockahead which was released prior
423 * to landing in master. This support will be dropped when 2.13 development
424 * starts. At the point, we should not just drop the connect flag (below), we
425 * should also remove the support in the code.
427 * Removing it means a few things:
428 * 1. Remove this section here
429 * 2. Remove CEF_NONBLOCK in ll_file_lockahead()
430 * 3. Remove function exp_connect_lockahead_old
431 * 4. Remove LDLM_FL_LOCKAHEAD_OLD_RESERVED in lustre_dlm_flags.h
433 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 12, 50, 0)
434 data->ocd_connect_flags |= OBD_CONNECT_LOCKAHEAD_OLD;
437 data->ocd_connect_flags2 = OBD_CONNECT2_LOCKAHEAD;
439 if (!OBD_FAIL_CHECK(OBD_FAIL_OSC_CONNECT_GRANT_PARAM))
440 data->ocd_connect_flags |= OBD_CONNECT_GRANT_PARAM;
442 /* OBD_CONNECT_CKSUM should always be set, even if checksums are
443 * disabled by default, because it can still be enabled on the
444 * fly via /sys. As a consequence, we still need to come to an
445 * agreement on the supported algorithms at connect time
447 data->ocd_connect_flags |= OBD_CONNECT_CKSUM;
449 if (OBD_FAIL_CHECK(OBD_FAIL_OSC_CKSUM_ADLER_ONLY))
450 data->ocd_cksum_types = OBD_CKSUM_ADLER;
452 data->ocd_cksum_types = obd_cksum_types_supported_client();
454 #ifdef HAVE_LRU_RESIZE_SUPPORT
455 data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
457 /* always ping even if server suppress_pings */
458 if (sbi->ll_flags & LL_SBI_ALWAYS_PING)
459 data->ocd_connect_flags &= ~OBD_CONNECT_PINGLESS;
461 CDEBUG(D_RPCTRACE, "ocd_connect_flags: %#llx ocd_version: %d "
462 "ocd_grant: %d\n", data->ocd_connect_flags,
463 data->ocd_version, data->ocd_grant);
465 sbi->ll_dt_obd->obd_upcall.onu_owner = &sbi->ll_lco;
466 sbi->ll_dt_obd->obd_upcall.onu_upcall = cl_ocd_update;
468 data->ocd_brw_size = DT_MAX_BRW_SIZE;
470 err = obd_connect(NULL, &sbi->ll_dt_exp, sbi->ll_dt_obd,
471 &sbi->ll_sb_uuid, data, sbi->ll_cache);
473 LCONSOLE_ERROR_MSG(0x150, "An OST (dt %s) is performing "
474 "recovery, of which this client is not a "
475 "part. Please wait for recovery to "
476 "complete, abort, or time out.\n", dt);
479 CERROR("%s: Cannot connect to %s: rc = %d\n",
480 sbi->ll_dt_exp->exp_obd->obd_name, dt, err);
484 sbi->ll_dt_exp->exp_connect_data = *data;
486 /* Don't change value if it was specified in the config log */
487 if (sbi->ll_ra_info.ra_max_read_ahead_whole_pages == -1)
488 sbi->ll_ra_info.ra_max_read_ahead_whole_pages =
489 max_t(unsigned long, SBI_DEFAULT_READAHEAD_WHOLE_MAX,
490 (data->ocd_brw_size >> PAGE_SHIFT));
492 err = obd_fid_init(sbi->ll_dt_exp->exp_obd, sbi->ll_dt_exp,
493 LUSTRE_SEQ_METADATA);
495 CERROR("%s: Can't init data layer FID infrastructure, "
496 "rc = %d\n", sbi->ll_dt_exp->exp_obd->obd_name, err);
500 mutex_lock(&sbi->ll_lco.lco_lock);
501 sbi->ll_lco.lco_flags = data->ocd_connect_flags;
502 sbi->ll_lco.lco_md_exp = sbi->ll_md_exp;
503 sbi->ll_lco.lco_dt_exp = sbi->ll_dt_exp;
504 mutex_unlock(&sbi->ll_lco.lco_lock);
506 fid_zero(&sbi->ll_root_fid);
507 err = md_get_root(sbi->ll_md_exp, get_mount_fileset(sb),
510 CERROR("cannot mds_connect: rc = %d\n", err);
511 GOTO(out_lock_cn_cb, err);
513 if (!fid_is_sane(&sbi->ll_root_fid)) {
514 CERROR("%s: Invalid root fid "DFID" during mount\n",
515 sbi->ll_md_exp->exp_obd->obd_name,
516 PFID(&sbi->ll_root_fid));
517 GOTO(out_lock_cn_cb, err = -EINVAL);
519 CDEBUG(D_SUPER, "rootfid "DFID"\n", PFID(&sbi->ll_root_fid));
521 sb->s_op = &lustre_super_operations;
522 #ifdef HAVE_XATTR_HANDLER_FLAGS
523 sb->s_xattr = ll_xattr_handlers;
525 #if THREAD_SIZE >= 8192 /*b=17630*/
526 sb->s_export_op = &lustre_export_operations;
530 * XXX: move this to after cbd setup? */
531 valid = OBD_MD_FLGETATTR | OBD_MD_FLBLOCKS | OBD_MD_FLMODEASIZE;
532 if (sbi->ll_flags & LL_SBI_ACL)
533 valid |= OBD_MD_FLACL;
535 OBD_ALLOC_PTR(op_data);
537 GOTO(out_lock_cn_cb, err = -ENOMEM);
539 op_data->op_fid1 = sbi->ll_root_fid;
540 op_data->op_mode = 0;
541 op_data->op_valid = valid;
543 err = md_getattr(sbi->ll_md_exp, op_data, &request);
545 OBD_FREE_PTR(op_data);
547 CERROR("%s: md_getattr failed for root: rc = %d\n",
548 sbi->ll_md_exp->exp_obd->obd_name, err);
549 GOTO(out_lock_cn_cb, err);
552 err = md_get_lustre_md(sbi->ll_md_exp, request, sbi->ll_dt_exp,
553 sbi->ll_md_exp, &lmd);
555 CERROR("failed to understand root inode md: rc = %d\n", err);
556 ptlrpc_req_finished(request);
557 GOTO(out_lock_cn_cb, err);
560 LASSERT(fid_is_sane(&sbi->ll_root_fid));
561 root = ll_iget(sb, cl_fid_build_ino(&sbi->ll_root_fid,
562 sbi->ll_flags & LL_SBI_32BIT_API),
564 md_free_lustre_md(sbi->ll_md_exp, &lmd);
565 ptlrpc_req_finished(request);
568 #ifdef CONFIG_FS_POSIX_ACL
570 posix_acl_release(lmd.posix_acl);
571 lmd.posix_acl = NULL;
574 err = IS_ERR(root) ? PTR_ERR(root) : -EBADF;
576 CERROR("lustre_lite: bad iget4 for root\n");
580 checksum = sbi->ll_flags & LL_SBI_CHECKSUM;
581 if (sbi->ll_checksum_set) {
582 err = obd_set_info_async(NULL, sbi->ll_dt_exp,
583 sizeof(KEY_CHECKSUM), KEY_CHECKSUM,
584 sizeof(checksum), &checksum, NULL);
586 CERROR("%s: Set checksum failed: rc = %d\n",
587 sbi->ll_dt_exp->exp_obd->obd_name, err);
593 sb->s_root = d_make_root(root);
594 if (sb->s_root == NULL) {
596 CERROR("%s: can't make root dentry: rc = %d\n",
597 sbi->ll_fsname, err);
600 #ifdef HAVE_DCACHE_LOCK
601 sb->s_root->d_op = &ll_d_ops;
604 sbi->ll_sdev_orig = sb->s_dev;
606 /* We set sb->s_dev equal on all lustre clients in order to support
607 * NFS export clustering. NFSD requires that the FSID be the same
609 /* s_dev is also used in lt_compare() to compare two fs, but that is
610 * only a node-local comparison. */
611 uuid = obd_get_uuid(sbi->ll_md_exp);
613 sb->s_dev = get_uuid2int(uuid->uuid, strlen(uuid->uuid));
620 if (sbi->ll_dt_obd) {
621 err = sysfs_create_link(&sbi->ll_kset.kobj,
622 &sbi->ll_dt_obd->obd_kset.kobj,
623 sbi->ll_dt_obd->obd_type->typ_name);
625 CERROR("%s: could not register %s in llite: rc = %d\n",
626 dt, sbi->ll_fsname, err);
631 if (sbi->ll_md_obd) {
632 err = sysfs_create_link(&sbi->ll_kset.kobj,
633 &sbi->ll_md_obd->obd_kset.kobj,
634 sbi->ll_md_obd->obd_type->typ_name);
636 CERROR("%s: could not register %s in llite: rc = %d\n",
637 md, sbi->ll_fsname, err);
647 obd_fid_fini(sbi->ll_dt_exp->exp_obd);
649 obd_disconnect(sbi->ll_dt_exp);
650 sbi->ll_dt_exp = NULL;
651 sbi->ll_dt_obd = NULL;
653 obd_fid_fini(sbi->ll_md_exp->exp_obd);
655 obd_disconnect(sbi->ll_md_exp);
656 sbi->ll_md_exp = NULL;
657 sbi->ll_md_obd = NULL;
666 int ll_get_max_mdsize(struct ll_sb_info *sbi, int *lmmsize)
670 size = sizeof(*lmmsize);
671 rc = obd_get_info(NULL, sbi->ll_dt_exp, sizeof(KEY_MAX_EASIZE),
672 KEY_MAX_EASIZE, &size, lmmsize);
674 CERROR("%s: cannot get max LOV EA size: rc = %d\n",
675 sbi->ll_dt_exp->exp_obd->obd_name, rc);
679 CDEBUG(D_INFO, "max LOV ea size: %d\n", *lmmsize);
682 rc = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_MAX_EASIZE),
683 KEY_MAX_EASIZE, &size, lmmsize);
685 CERROR("Get max mdsize error rc %d\n", rc);
687 CDEBUG(D_INFO, "max LMV ea size: %d\n", *lmmsize);
693 * Get the value of the default_easize parameter.
695 * \see client_obd::cl_default_mds_easize
697 * \param[in] sbi superblock info for this filesystem
698 * \param[out] lmmsize pointer to storage location for value
700 * \retval 0 on success
701 * \retval negative negated errno on failure
703 int ll_get_default_mdsize(struct ll_sb_info *sbi, int *lmmsize)
708 rc = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_DEFAULT_EASIZE),
709 KEY_DEFAULT_EASIZE, &size, lmmsize);
711 CERROR("Get default mdsize error rc %d\n", rc);
717 * Set the default_easize parameter to the given value.
719 * \see client_obd::cl_default_mds_easize
721 * \param[in] sbi superblock info for this filesystem
722 * \param[in] lmmsize the size to set
724 * \retval 0 on success
725 * \retval negative negated errno on failure
727 int ll_set_default_mdsize(struct ll_sb_info *sbi, int lmmsize)
731 if (lmmsize < sizeof(struct lov_mds_md) ||
732 lmmsize > OBD_MAX_DEFAULT_EA_SIZE)
735 rc = obd_set_info_async(NULL, sbi->ll_md_exp,
736 sizeof(KEY_DEFAULT_EASIZE), KEY_DEFAULT_EASIZE,
737 sizeof(int), &lmmsize, NULL);
742 static void client_common_put_super(struct super_block *sb)
744 struct ll_sb_info *sbi = ll_s2sbi(sb);
749 obd_fid_fini(sbi->ll_dt_exp->exp_obd);
750 obd_disconnect(sbi->ll_dt_exp);
751 sbi->ll_dt_exp = NULL;
753 ll_debugfs_unregister_super(sb);
755 obd_fid_fini(sbi->ll_md_exp->exp_obd);
756 obd_disconnect(sbi->ll_md_exp);
757 sbi->ll_md_exp = NULL;
762 void ll_kill_super(struct super_block *sb)
764 struct ll_sb_info *sbi;
768 if (!(sb->s_flags & MS_ACTIVE))
772 /* we need restore s_dev from changed for clustred NFS before put_super
773 * because new kernels have cached s_dev and change sb->s_dev in
774 * put_super not affected real removing devices */
776 sb->s_dev = sbi->ll_sdev_orig;
777 sbi->ll_umounting = 1;
779 /* wait running statahead threads to quit */
780 while (atomic_read(&sbi->ll_sa_running) > 0) {
781 set_current_state(TASK_UNINTERRUPTIBLE);
782 schedule_timeout(msecs_to_jiffies(MSEC_PER_SEC >> 3));
789 static inline int ll_set_opt(const char *opt, char *data, int fl)
791 if (strncmp(opt, data, strlen(opt)) != 0)
797 /* non-client-specific mount options are parsed in lmd_parse */
798 static int ll_options(char *options, struct ll_sb_info *sbi)
801 char *s1 = options, *s2;
802 int *flags = &sbi->ll_flags;
808 CDEBUG(D_CONFIG, "Parsing opts %s\n", options);
811 CDEBUG(D_SUPER, "next opt=%s\n", s1);
812 tmp = ll_set_opt("nolock", s1, LL_SBI_NOLCK);
817 tmp = ll_set_opt("flock", s1, LL_SBI_FLOCK);
822 tmp = ll_set_opt("localflock", s1, LL_SBI_LOCALFLOCK);
827 tmp = ll_set_opt("noflock", s1, LL_SBI_FLOCK|LL_SBI_LOCALFLOCK);
832 tmp = ll_set_opt("user_xattr", s1, LL_SBI_USER_XATTR);
837 tmp = ll_set_opt("nouser_xattr", s1, LL_SBI_USER_XATTR);
842 tmp = ll_set_opt("context", s1, 1);
845 tmp = ll_set_opt("fscontext", s1, 1);
848 tmp = ll_set_opt("defcontext", s1, 1);
851 tmp = ll_set_opt("rootcontext", s1, 1);
854 tmp = ll_set_opt("user_fid2path", s1, LL_SBI_USER_FID2PATH);
859 tmp = ll_set_opt("nouser_fid2path", s1, LL_SBI_USER_FID2PATH);
865 tmp = ll_set_opt("checksum", s1, LL_SBI_CHECKSUM);
868 sbi->ll_checksum_set = 1;
871 tmp = ll_set_opt("nochecksum", s1, LL_SBI_CHECKSUM);
874 sbi->ll_checksum_set = 1;
877 tmp = ll_set_opt("lruresize", s1, LL_SBI_LRU_RESIZE);
882 tmp = ll_set_opt("nolruresize", s1, LL_SBI_LRU_RESIZE);
887 tmp = ll_set_opt("lazystatfs", s1, LL_SBI_LAZYSTATFS);
892 tmp = ll_set_opt("nolazystatfs", s1, LL_SBI_LAZYSTATFS);
897 tmp = ll_set_opt("32bitapi", s1, LL_SBI_32BIT_API);
902 tmp = ll_set_opt("verbose", s1, LL_SBI_VERBOSE);
907 tmp = ll_set_opt("noverbose", s1, LL_SBI_VERBOSE);
912 tmp = ll_set_opt("always_ping", s1, LL_SBI_ALWAYS_PING);
917 LCONSOLE_ERROR_MSG(0x152, "Unknown option '%s', won't mount.\n",
923 s2 = strchr(s1, ',');
931 void ll_lli_init(struct ll_inode_info *lli)
933 lli->lli_inode_magic = LLI_INODE_MAGIC;
935 spin_lock_init(&lli->lli_lock);
936 lli->lli_posix_acl = NULL;
937 /* Do not set lli_fid, it has been initialized already. */
938 fid_zero(&lli->lli_pfid);
939 lli->lli_mds_read_och = NULL;
940 lli->lli_mds_write_och = NULL;
941 lli->lli_mds_exec_och = NULL;
942 lli->lli_open_fd_read_count = 0;
943 lli->lli_open_fd_write_count = 0;
944 lli->lli_open_fd_exec_count = 0;
945 mutex_init(&lli->lli_och_mutex);
946 spin_lock_init(&lli->lli_agl_lock);
947 spin_lock_init(&lli->lli_layout_lock);
948 ll_layout_version_set(lli, CL_LAYOUT_GEN_NONE);
949 lli->lli_clob = NULL;
951 init_rwsem(&lli->lli_xattrs_list_rwsem);
952 mutex_init(&lli->lli_xattrs_enq_lock);
954 LASSERT(lli->lli_vfs_inode.i_mode != 0);
955 if (S_ISDIR(lli->lli_vfs_inode.i_mode)) {
956 mutex_init(&lli->lli_readdir_mutex);
957 lli->lli_opendir_key = NULL;
959 spin_lock_init(&lli->lli_sa_lock);
960 lli->lli_opendir_pid = 0;
961 lli->lli_sa_enabled = 0;
962 lli->lli_def_stripe_offset = -1;
963 init_rwsem(&lli->lli_lsm_sem);
965 mutex_init(&lli->lli_size_mutex);
966 lli->lli_symlink_name = NULL;
967 init_rwsem(&lli->lli_trunc_sem);
968 range_lock_tree_init(&lli->lli_write_tree);
969 init_rwsem(&lli->lli_glimpse_sem);
970 lli->lli_glimpse_time = ktime_set(0, 0);
971 INIT_LIST_HEAD(&lli->lli_agl_list);
972 lli->lli_agl_index = 0;
973 lli->lli_async_rc = 0;
974 spin_lock_init(&lli->lli_heat_lock);
975 obd_heat_clear(lli->lli_heat_instances, OBD_HEAT_COUNT);
976 lli->lli_heat_flags = 0;
978 mutex_init(&lli->lli_layout_mutex);
979 memset(lli->lli_jobid, 0, sizeof(lli->lli_jobid));
982 #define MAX_STRING_SIZE 128
984 #ifndef HAVE_SUPER_SETUP_BDI_NAME
986 #define LSI_BDI_INITIALIZED 0x00400000
988 #ifndef HAVE_BDI_CAP_MAP_COPY
989 # define BDI_CAP_MAP_COPY 0
992 static int super_setup_bdi_name(struct super_block *sb, char *fmt, ...)
994 struct lustre_sb_info *lsi = s2lsi(sb);
995 char buf[MAX_STRING_SIZE];
999 err = bdi_init(&lsi->lsi_bdi);
1003 lsi->lsi_flags |= LSI_BDI_INITIALIZED;
1004 lsi->lsi_bdi.capabilities = BDI_CAP_MAP_COPY;
1005 lsi->lsi_bdi.name = "lustre";
1006 va_start(args, fmt);
1007 vsnprintf(buf, MAX_STRING_SIZE, fmt, args);
1009 err = bdi_register(&lsi->lsi_bdi, NULL, "%s", buf);
1012 sb->s_bdi = &lsi->lsi_bdi;
1016 #endif /* !HAVE_SUPER_SETUP_BDI_NAME */
1018 int ll_fill_super(struct super_block *sb, struct vfsmount *mnt)
1020 struct lustre_profile *lprof = NULL;
1021 struct lustre_sb_info *lsi = s2lsi(sb);
1022 struct ll_sb_info *sbi = NULL;
1023 char *dt = NULL, *md = NULL;
1024 char *profilenm = get_profile_name(sb);
1025 struct config_llog_instance *cfg;
1026 /* %p for void* in printf needs 16+2 characters: 0xffffffffffffffff */
1027 const int instlen = 16 + 2;
1028 unsigned long cfg_instance = ll_get_cfg_instance(sb);
1029 char name[MAX_STRING_SIZE];
1038 /* for ASLR, to map between cfg_instance and hashed ptr */
1039 CDEBUG(D_VFSTRACE, "VFS Op: cfg_instance %s-%016lx (sb %p)\n",
1040 profilenm, cfg_instance, sb);
1042 try_module_get(THIS_MODULE);
1046 GOTO(out_free_cfg, err = -ENOMEM);
1048 /* client additional sb info */
1049 lsi->lsi_llsbi = sbi = ll_init_sbi();
1051 GOTO(out_free_cfg, err = -ENOMEM);
1053 err = ll_options(lsi->lsi_lmd->lmd_opts, sbi);
1055 GOTO(out_free_cfg, err);
1057 #ifndef HAVE_DCACHE_LOCK
1058 /* kernel >= 2.6.38 store dentry operations in sb->s_d_op. */
1059 sb->s_d_op = &ll_d_ops;
1062 generate_random_uuid(uuid.b);
1063 snprintf(sbi->ll_sb_uuid.uuid, UUID_SIZE, "%pU", uuid.b);
1065 CDEBUG(D_CONFIG, "llite sb uuid: %s\n", sbi->ll_sb_uuid.uuid);
1068 len = strlen(profilenm);
1069 ptr = strrchr(profilenm, '-');
1070 if (ptr && (strcmp(ptr, "-client") == 0))
1073 if (len > LUSTRE_MAXFSNAME) {
1074 if (unlikely(len >= MAX_STRING_SIZE))
1075 len = MAX_STRING_SIZE - 1;
1076 strncpy(name, profilenm, len);
1078 err = -ENAMETOOLONG;
1079 CERROR("%s: fsname longer than %u characters: rc = %d\n",
1080 name, LUSTRE_MAXFSNAME, err);
1081 GOTO(out_free_cfg, err);
1083 strncpy(sbi->ll_fsname, profilenm, len);
1084 sbi->ll_fsname[len] = '\0';
1087 snprintf(name, sizeof(name), "%.*s-%016lx", len,
1088 profilenm, cfg_instance);
1090 err = super_setup_bdi_name(sb, "%s", name);
1092 GOTO(out_free_cfg, err);
1094 /* Call ll_debugfs_register_super() before lustre_process_log()
1095 * so that "llite.*.*" params can be processed correctly.
1097 err = ll_debugfs_register_super(sb, name);
1099 CERROR("%s: could not register mountpoint in llite: rc = %d\n",
1100 sbi->ll_fsname, err);
1104 /* The cfg_instance is a value unique to this super, in case some
1105 * joker tries to mount the same fs at two mount points.
1107 cfg->cfg_instance = cfg_instance;
1108 cfg->cfg_uuid = lsi->lsi_llsbi->ll_sb_uuid;
1109 cfg->cfg_callback = class_config_llog_handler;
1110 cfg->cfg_sub_clds = CONFIG_SUB_CLIENT;
1111 /* set up client obds */
1112 err = lustre_process_log(sb, profilenm, cfg);
1114 GOTO(out_debugfs, err);
1116 /* Profile set with LCFG_MOUNTOPT so we can find our mdc and osc obds */
1117 lprof = class_get_profile(profilenm);
1118 if (lprof == NULL) {
1119 LCONSOLE_ERROR_MSG(0x156, "The client profile '%s' could not be"
1120 " read from the MGS. Does that filesystem "
1121 "exist?\n", profilenm);
1122 GOTO(out_debugfs, err = -EINVAL);
1124 CDEBUG(D_CONFIG, "Found profile %s: mdc=%s osc=%s\n", profilenm,
1125 lprof->lp_md, lprof->lp_dt);
1127 dt_len = strlen(lprof->lp_dt) + instlen + 2;
1128 OBD_ALLOC(dt, dt_len);
1130 GOTO(out_profile, err = -ENOMEM);
1131 snprintf(dt, dt_len - 1, "%s-%016lx", lprof->lp_dt, cfg_instance);
1133 md_len = strlen(lprof->lp_md) + instlen + 2;
1134 OBD_ALLOC(md, md_len);
1136 GOTO(out_free_dt, err = -ENOMEM);
1137 snprintf(md, md_len - 1, "%s-%016lx", lprof->lp_md, cfg_instance);
1139 /* connections, registrations, sb setup */
1140 err = client_common_fill_super(sb, md, dt, mnt);
1142 GOTO(out_free_md, err);
1144 sbi->ll_client_common_fill_super_succeeded = 1;
1148 OBD_FREE(md, md_len);
1151 OBD_FREE(dt, dt_len);
1154 class_put_profile(lprof);
1157 ll_debugfs_unregister_super(sb);
1164 else if (sbi->ll_flags & LL_SBI_VERBOSE)
1165 LCONSOLE_WARN("Mounted %s\n", profilenm);
1167 } /* ll_fill_super */
1169 void ll_put_super(struct super_block *sb)
1171 struct config_llog_instance cfg, params_cfg;
1172 struct obd_device *obd;
1173 struct lustre_sb_info *lsi = s2lsi(sb);
1174 struct ll_sb_info *sbi = ll_s2sbi(sb);
1175 char *profilenm = get_profile_name(sb);
1176 unsigned long cfg_instance = ll_get_cfg_instance(sb);
1178 int next, force = 1, rc = 0;
1182 GOTO(out_no_sbi, 0);
1184 /* Should replace instance_id with something better for ASLR */
1185 CDEBUG(D_VFSTRACE, "VFS Op: cfg_instance %s-%016lx (sb %p)\n",
1186 profilenm, cfg_instance, sb);
1188 cfg.cfg_instance = cfg_instance;
1189 lustre_end_log(sb, profilenm, &cfg);
1191 params_cfg.cfg_instance = cfg_instance;
1192 lustre_end_log(sb, PARAMS_FILENAME, ¶ms_cfg);
1194 if (sbi->ll_md_exp) {
1195 obd = class_exp2obd(sbi->ll_md_exp);
1197 force = obd->obd_force;
1200 /* Wait for unstable pages to be committed to stable storage */
1202 struct l_wait_info lwi = LWI_INTR(LWI_ON_SIGNAL_NOOP, NULL);
1203 rc = l_wait_event(sbi->ll_cache->ccc_unstable_waitq,
1204 atomic_long_read(&sbi->ll_cache->ccc_unstable_nr) == 0,
1208 ccc_count = atomic_long_read(&sbi->ll_cache->ccc_unstable_nr);
1209 if (force == 0 && rc != -EINTR)
1210 LASSERTF(ccc_count == 0, "count: %li\n", ccc_count);
1212 /* We need to set force before the lov_disconnect in
1213 lustre_common_put_super, since l_d cleans up osc's as well. */
1216 while ((obd = class_devices_in_group(&sbi->ll_sb_uuid,
1218 obd->obd_force = force;
1222 if (sbi->ll_client_common_fill_super_succeeded) {
1223 /* Only if client_common_fill_super succeeded */
1224 client_common_put_super(sb);
1228 while ((obd = class_devices_in_group(&sbi->ll_sb_uuid, &next)) !=NULL) {
1229 class_manual_cleanup(obd);
1232 if (sbi->ll_flags & LL_SBI_VERBOSE)
1233 LCONSOLE_WARN("Unmounted %s\n", profilenm ? profilenm : "");
1236 class_del_profile(profilenm);
1238 #ifndef HAVE_SUPER_SETUP_BDI_NAME
1239 if (lsi->lsi_flags & LSI_BDI_INITIALIZED) {
1240 bdi_destroy(&lsi->lsi_bdi);
1241 lsi->lsi_flags &= ~LSI_BDI_INITIALIZED;
1246 lsi->lsi_llsbi = NULL;
1248 lustre_common_put_super(sb);
1250 cl_env_cache_purge(~0);
1252 module_put(THIS_MODULE);
1255 } /* client_put_super */
1257 struct inode *ll_inode_from_resource_lock(struct ldlm_lock *lock)
1259 struct inode *inode = NULL;
1261 /* NOTE: we depend on atomic igrab() -bzzz */
1262 lock_res_and_lock(lock);
1263 if (lock->l_resource->lr_lvb_inode) {
1264 struct ll_inode_info * lli;
1265 lli = ll_i2info(lock->l_resource->lr_lvb_inode);
1266 if (lli->lli_inode_magic == LLI_INODE_MAGIC) {
1267 inode = igrab(lock->l_resource->lr_lvb_inode);
1269 inode = lock->l_resource->lr_lvb_inode;
1270 LDLM_DEBUG_LIMIT(inode->i_state & I_FREEING ? D_INFO :
1271 D_WARNING, lock, "lr_lvb_inode %p is "
1272 "bogus: magic %08x",
1273 lock->l_resource->lr_lvb_inode,
1274 lli->lli_inode_magic);
1278 unlock_res_and_lock(lock);
1282 void ll_dir_clear_lsm_md(struct inode *inode)
1284 struct ll_inode_info *lli = ll_i2info(inode);
1286 LASSERT(S_ISDIR(inode->i_mode));
1288 if (lli->lli_lsm_md != NULL) {
1289 lmv_free_memmd(lli->lli_lsm_md);
1290 lli->lli_lsm_md = NULL;
1294 static struct inode *ll_iget_anon_dir(struct super_block *sb,
1295 const struct lu_fid *fid,
1296 struct lustre_md *md)
1298 struct ll_sb_info *sbi = ll_s2sbi(sb);
1299 struct mdt_body *body = md->body;
1300 struct inode *inode;
1304 ino = cl_fid_build_ino(fid, sbi->ll_flags & LL_SBI_32BIT_API);
1305 inode = iget_locked(sb, ino);
1306 if (inode == NULL) {
1307 CERROR("%s: failed get simple inode "DFID": rc = -ENOENT\n",
1308 sbi->ll_fsname, PFID(fid));
1309 RETURN(ERR_PTR(-ENOENT));
1312 if (inode->i_state & I_NEW) {
1313 struct ll_inode_info *lli = ll_i2info(inode);
1314 struct lmv_stripe_md *lsm = md->lmv;
1316 inode->i_mode = (inode->i_mode & ~S_IFMT) |
1317 (body->mbo_mode & S_IFMT);
1318 LASSERTF(S_ISDIR(inode->i_mode), "Not slave inode "DFID"\n",
1321 inode->i_mtime.tv_sec = 0;
1322 inode->i_atime.tv_sec = 0;
1323 inode->i_ctime.tv_sec = 0;
1326 #ifdef HAVE_BACKING_DEV_INFO
1327 /* initializing backing dev info. */
1328 inode->i_mapping->backing_dev_info =
1329 &s2lsi(inode->i_sb)->lsi_bdi;
1331 inode->i_op = &ll_dir_inode_operations;
1332 inode->i_fop = &ll_dir_operations;
1333 lli->lli_fid = *fid;
1336 LASSERT(lsm != NULL);
1337 /* master object FID */
1338 lli->lli_pfid = body->mbo_fid1;
1339 CDEBUG(D_INODE, "lli %p slave "DFID" master "DFID"\n",
1340 lli, PFID(fid), PFID(&lli->lli_pfid));
1341 unlock_new_inode(inode);
1347 static int ll_init_lsm_md(struct inode *inode, struct lustre_md *md)
1350 struct lmv_stripe_md *lsm = md->lmv;
1351 struct ll_inode_info *lli = ll_i2info(inode);
1354 LASSERT(lsm != NULL);
1356 CDEBUG(D_INODE, "%s: "DFID" set dir layout:\n",
1357 ll_i2sbi(inode)->ll_fsname, PFID(&lli->lli_fid));
1358 lsm_md_dump(D_INODE, lsm);
1360 /* XXX sigh, this lsm_root initialization should be in
1361 * LMV layer, but it needs ll_iget right now, so we
1362 * put this here right now. */
1363 for (i = 0; i < lsm->lsm_md_stripe_count; i++) {
1364 fid = &lsm->lsm_md_oinfo[i].lmo_fid;
1365 LASSERT(lsm->lsm_md_oinfo[i].lmo_root == NULL);
1367 if (!fid_is_sane(fid))
1370 /* Unfortunately ll_iget will call ll_update_inode,
1371 * where the initialization of slave inode is slightly
1372 * different, so it reset lsm_md to NULL to avoid
1373 * initializing lsm for slave inode. */
1374 lsm->lsm_md_oinfo[i].lmo_root =
1375 ll_iget_anon_dir(inode->i_sb, fid, md);
1376 if (IS_ERR(lsm->lsm_md_oinfo[i].lmo_root)) {
1377 int rc = PTR_ERR(lsm->lsm_md_oinfo[i].lmo_root);
1379 lsm->lsm_md_oinfo[i].lmo_root = NULL;
1381 iput(lsm->lsm_md_oinfo[i].lmo_root);
1382 lsm->lsm_md_oinfo[i].lmo_root = NULL;
1388 lli->lli_lsm_md = lsm;
1393 static int ll_update_lsm_md(struct inode *inode, struct lustre_md *md)
1395 struct ll_inode_info *lli = ll_i2info(inode);
1396 struct lmv_stripe_md *lsm = md->lmv;
1401 LASSERT(S_ISDIR(inode->i_mode));
1402 CDEBUG(D_INODE, "update lsm %p of "DFID"\n", lli->lli_lsm_md,
1403 PFID(ll_inode2fid(inode)));
1406 * no striped information from request, lustre_md from req does not
1407 * include stripeEA, see ll_md_setattr()
1413 * normally dir layout doesn't change, only take read lock to check
1414 * that to avoid blocking other MD operations.
1416 if (lli->lli_lsm_md)
1417 down_read(&lli->lli_lsm_sem);
1419 down_write(&lli->lli_lsm_sem);
1422 * if dir layout mismatch, check whether version is increased, which
1423 * means layout is changed, this happens in dir migration and lfsck.
1425 * foreign LMV should not change.
1427 if (lli->lli_lsm_md &&
1428 lli->lli_lsm_md->lsm_md_magic != LMV_MAGIC_FOREIGN &&
1429 !lsm_md_eq(lli->lli_lsm_md, lsm)) {
1430 if (lsm->lsm_md_layout_version <=
1431 lli->lli_lsm_md->lsm_md_layout_version) {
1432 CERROR("%s: "DFID" dir layout mismatch:\n",
1433 ll_i2sbi(inode)->ll_fsname,
1434 PFID(&lli->lli_fid));
1435 lsm_md_dump(D_ERROR, lli->lli_lsm_md);
1436 lsm_md_dump(D_ERROR, lsm);
1437 GOTO(unlock, rc = -EINVAL);
1440 /* layout changed, switch to write lock */
1441 up_read(&lli->lli_lsm_sem);
1442 down_write(&lli->lli_lsm_sem);
1443 ll_dir_clear_lsm_md(inode);
1446 /* set directory layout */
1447 if (!lli->lli_lsm_md) {
1448 struct cl_attr *attr;
1450 if (lsm->lsm_md_magic == LMV_MAGIC_FOREIGN) {
1451 /* set md->lmv to NULL, so the following free lustre_md
1452 * will not free this lsm */
1454 lli->lli_lsm_md = lsm;
1455 up_write(&lli->lli_lsm_sem);
1459 rc = ll_init_lsm_md(inode, md);
1460 up_write(&lli->lli_lsm_sem);
1464 /* set md->lmv to NULL, so the following free lustre_md
1465 * will not free this lsm */
1469 * md_merge_attr() may take long, since lsm is already set,
1470 * switch to read lock.
1472 down_read(&lli->lli_lsm_sem);
1474 OBD_ALLOC_PTR(attr);
1476 GOTO(unlock, rc = -ENOMEM);
1478 /* validate the lsm */
1479 rc = md_merge_attr(ll_i2mdexp(inode), lsm, attr,
1480 ll_md_blocking_ast);
1486 if (md->body->mbo_valid & OBD_MD_FLNLINK)
1487 md->body->mbo_nlink = attr->cat_nlink;
1488 if (md->body->mbo_valid & OBD_MD_FLSIZE)
1489 md->body->mbo_size = attr->cat_size;
1490 if (md->body->mbo_valid & OBD_MD_FLATIME)
1491 md->body->mbo_atime = attr->cat_atime;
1492 if (md->body->mbo_valid & OBD_MD_FLCTIME)
1493 md->body->mbo_ctime = attr->cat_ctime;
1494 if (md->body->mbo_valid & OBD_MD_FLMTIME)
1495 md->body->mbo_mtime = attr->cat_mtime;
1500 up_read(&lli->lli_lsm_sem);
1505 void ll_clear_inode(struct inode *inode)
1507 struct ll_inode_info *lli = ll_i2info(inode);
1508 struct ll_sb_info *sbi = ll_i2sbi(inode);
1511 CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p)\n",
1512 PFID(ll_inode2fid(inode)), inode);
1514 if (S_ISDIR(inode->i_mode)) {
1515 /* these should have been cleared in ll_file_release */
1516 LASSERT(lli->lli_opendir_key == NULL);
1517 LASSERT(lli->lli_sai == NULL);
1518 LASSERT(lli->lli_opendir_pid == 0);
1521 md_null_inode(sbi->ll_md_exp, ll_inode2fid(inode));
1523 LASSERT(!lli->lli_open_fd_write_count);
1524 LASSERT(!lli->lli_open_fd_read_count);
1525 LASSERT(!lli->lli_open_fd_exec_count);
1527 if (lli->lli_mds_write_och)
1528 ll_md_real_close(inode, FMODE_WRITE);
1529 if (lli->lli_mds_exec_och)
1530 ll_md_real_close(inode, FMODE_EXEC);
1531 if (lli->lli_mds_read_och)
1532 ll_md_real_close(inode, FMODE_READ);
1534 if (S_ISLNK(inode->i_mode) && lli->lli_symlink_name) {
1535 OBD_FREE(lli->lli_symlink_name,
1536 strlen(lli->lli_symlink_name) + 1);
1537 lli->lli_symlink_name = NULL;
1540 ll_xattr_cache_destroy(inode);
1542 #ifdef CONFIG_FS_POSIX_ACL
1543 forget_all_cached_acls(inode);
1544 if (lli->lli_posix_acl) {
1545 posix_acl_release(lli->lli_posix_acl);
1546 lli->lli_posix_acl = NULL;
1549 lli->lli_inode_magic = LLI_INODE_DEAD;
1551 if (S_ISDIR(inode->i_mode))
1552 ll_dir_clear_lsm_md(inode);
1553 else if (S_ISREG(inode->i_mode) && !is_bad_inode(inode))
1554 LASSERT(list_empty(&lli->lli_agl_list));
1557 * XXX This has to be done before lsm is freed below, because
1558 * cl_object still uses inode lsm.
1560 cl_inode_fini(inode);
1565 static int ll_md_setattr(struct dentry *dentry, struct md_op_data *op_data)
1567 struct lustre_md md;
1568 struct inode *inode = dentry->d_inode;
1569 struct ll_sb_info *sbi = ll_i2sbi(inode);
1570 struct ptlrpc_request *request = NULL;
1574 op_data = ll_prep_md_op_data(op_data, inode, NULL, NULL, 0, 0,
1575 LUSTRE_OPC_ANY, NULL);
1576 if (IS_ERR(op_data))
1577 RETURN(PTR_ERR(op_data));
1579 rc = md_setattr(sbi->ll_md_exp, op_data, NULL, 0, &request);
1581 ptlrpc_req_finished(request);
1582 if (rc == -ENOENT) {
1584 /* Unlinked special device node? Or just a race?
1585 * Pretend we done everything. */
1586 if (!S_ISREG(inode->i_mode) &&
1587 !S_ISDIR(inode->i_mode)) {
1588 ia_valid = op_data->op_attr.ia_valid;
1589 op_data->op_attr.ia_valid &= ~TIMES_SET_FLAGS;
1590 rc = simple_setattr(dentry, &op_data->op_attr);
1591 op_data->op_attr.ia_valid = ia_valid;
1593 } else if (rc != -EPERM && rc != -EACCES && rc != -ETXTBSY) {
1594 CERROR("md_setattr fails: rc = %d\n", rc);
1599 rc = md_get_lustre_md(sbi->ll_md_exp, request, sbi->ll_dt_exp,
1600 sbi->ll_md_exp, &md);
1602 ptlrpc_req_finished(request);
1606 ia_valid = op_data->op_attr.ia_valid;
1607 /* inode size will be in ll_setattr_ost, can't do it now since dirty
1608 * cache is not cleared yet. */
1609 op_data->op_attr.ia_valid &= ~(TIMES_SET_FLAGS | ATTR_SIZE);
1610 if (S_ISREG(inode->i_mode))
1612 rc = simple_setattr(dentry, &op_data->op_attr);
1613 if (S_ISREG(inode->i_mode))
1614 inode_unlock(inode);
1615 op_data->op_attr.ia_valid = ia_valid;
1617 rc = ll_update_inode(inode, &md);
1618 ptlrpc_req_finished(request);
1623 /* If this inode has objects allocated to it (lsm != NULL), then the OST
1624 * object(s) determine the file size and mtime. Otherwise, the MDS will
1625 * keep these values until such a time that objects are allocated for it.
1626 * We do the MDS operations first, as it is checking permissions for us.
1627 * We don't to the MDS RPC if there is nothing that we want to store there,
1628 * otherwise there is no harm in updating mtime/atime on the MDS if we are
1629 * going to do an RPC anyways.
1631 * If we are doing a truncate, we will send the mtime and ctime updates
1632 * to the OST with the punch RPC, otherwise we do an explicit setattr RPC.
1633 * I don't believe it is possible to get e.g. ATTR_MTIME_SET and ATTR_SIZE
1636 * In case of HSMimport, we only set attr on MDS.
1638 int ll_setattr_raw(struct dentry *dentry, struct iattr *attr,
1639 enum op_xvalid xvalid, bool hsm_import)
1641 struct inode *inode = dentry->d_inode;
1642 struct ll_inode_info *lli = ll_i2info(inode);
1643 struct md_op_data *op_data = NULL;
1647 CDEBUG(D_VFSTRACE, "%s: setattr inode "DFID"(%p) from %llu to %llu, "
1648 "valid %x, hsm_import %d\n",
1649 ll_i2sbi(inode)->ll_fsname, PFID(&lli->lli_fid),
1650 inode, i_size_read(inode), attr->ia_size, attr->ia_valid,
1653 if (attr->ia_valid & ATTR_SIZE) {
1654 /* Check new size against VFS/VM file size limit and rlimit */
1655 rc = inode_newsize_ok(inode, attr->ia_size);
1659 /* The maximum Lustre file size is variable, based on the
1660 * OST maximum object size and number of stripes. This
1661 * needs another check in addition to the VFS check above. */
1662 if (attr->ia_size > ll_file_maxbytes(inode)) {
1663 CDEBUG(D_INODE,"file "DFID" too large %llu > %llu\n",
1664 PFID(&lli->lli_fid), attr->ia_size,
1665 ll_file_maxbytes(inode));
1669 attr->ia_valid |= ATTR_MTIME | ATTR_CTIME;
1672 /* POSIX: check before ATTR_*TIME_SET set (from inode_change_ok) */
1673 if (attr->ia_valid & TIMES_SET_FLAGS) {
1674 if ((!uid_eq(current_fsuid(), inode->i_uid)) &&
1675 !cfs_capable(CFS_CAP_FOWNER))
1679 /* We mark all of the fields "set" so MDS/OST does not re-set them */
1680 if (!(xvalid & OP_XVALID_CTIME_SET) &&
1681 (attr->ia_valid & ATTR_CTIME)) {
1682 attr->ia_ctime = current_time(inode);
1683 xvalid |= OP_XVALID_CTIME_SET;
1685 if (!(attr->ia_valid & ATTR_ATIME_SET) &&
1686 (attr->ia_valid & ATTR_ATIME)) {
1687 attr->ia_atime = current_time(inode);
1688 attr->ia_valid |= ATTR_ATIME_SET;
1690 if (!(attr->ia_valid & ATTR_MTIME_SET) &&
1691 (attr->ia_valid & ATTR_MTIME)) {
1692 attr->ia_mtime = current_time(inode);
1693 attr->ia_valid |= ATTR_MTIME_SET;
1696 if (attr->ia_valid & (ATTR_MTIME | ATTR_CTIME))
1697 CDEBUG(D_INODE, "setting mtime %lld, ctime %lld, now = %lld\n",
1698 (s64)attr->ia_mtime.tv_sec, (s64)attr->ia_ctime.tv_sec,
1699 ktime_get_real_seconds());
1701 if (S_ISREG(inode->i_mode)) {
1702 if (attr->ia_valid & ATTR_SIZE)
1703 inode_dio_write_done(inode);
1704 inode_unlock(inode);
1707 /* We always do an MDS RPC, even if we're only changing the size;
1708 * only the MDS knows whether truncate() should fail with -ETXTBUSY */
1710 OBD_ALLOC_PTR(op_data);
1711 if (op_data == NULL)
1712 GOTO(out, rc = -ENOMEM);
1714 if (!hsm_import && attr->ia_valid & ATTR_SIZE) {
1715 /* If we are changing file size, file content is
1716 * modified, flag it.
1718 xvalid |= OP_XVALID_OWNEROVERRIDE;
1719 op_data->op_bias |= MDS_DATA_MODIFIED;
1720 ll_file_clear_flag(lli, LLIF_DATA_MODIFIED);
1723 if (attr->ia_valid & ATTR_FILE) {
1724 struct ll_file_data *fd = LUSTRE_FPRIVATE(attr->ia_file);
1726 if (fd->fd_lease_och)
1727 op_data->op_bias |= MDS_TRUNC_KEEP_LEASE;
1730 op_data->op_attr = *attr;
1731 op_data->op_xvalid = xvalid;
1733 rc = ll_md_setattr(dentry, op_data);
1737 if (!S_ISREG(inode->i_mode) || hsm_import)
1740 if (attr->ia_valid & (ATTR_SIZE | ATTR_ATIME | ATTR_ATIME_SET |
1741 ATTR_MTIME | ATTR_MTIME_SET | ATTR_CTIME) ||
1742 xvalid & OP_XVALID_CTIME_SET) {
1743 /* For truncate and utimes sending attributes to OSTs, setting
1744 * mtime/atime to the past will be performed under PW [0:EOF]
1745 * extent lock (new_size:EOF for truncate). It may seem
1746 * excessive to send mtime/atime updates to OSTs when not
1747 * setting times to past, but it is necessary due to possible
1748 * time de-synchronization between MDT inode and OST objects
1750 rc = cl_setattr_ost(lli->lli_clob, attr, xvalid, 0);
1753 /* If the file was restored, it needs to set dirty flag.
1755 * We've already sent MDS_DATA_MODIFIED flag in
1756 * ll_md_setattr() for truncate. However, the MDT refuses to
1757 * set the HS_DIRTY flag on released files, so we have to set
1758 * it again if the file has been restored. Please check how
1759 * LLIF_DATA_MODIFIED is set in vvp_io_setattr_fini().
1761 * Please notice that if the file is not released, the previous
1762 * MDS_DATA_MODIFIED has taken effect and usually
1763 * LLIF_DATA_MODIFIED is not set(see vvp_io_setattr_fini()).
1764 * This way we can save an RPC for common open + trunc
1766 if (ll_file_test_and_clear_flag(lli, LLIF_DATA_MODIFIED)) {
1767 struct hsm_state_set hss = {
1768 .hss_valid = HSS_SETMASK,
1769 .hss_setmask = HS_DIRTY,
1773 rc2 = ll_hsm_state_set(inode, &hss);
1774 /* truncate and write can happen at the same time, so that
1775 * the file can be set modified even though the file is not
1776 * restored from released state, and ll_hsm_state_set() is
1777 * not applicable for the file, and rc2 < 0 is normal in this
1780 CDEBUG(D_INFO, DFID "HSM set dirty failed: rc2 = %d\n",
1781 PFID(ll_inode2fid(inode)), rc2);
1786 if (op_data != NULL)
1787 ll_finish_md_op_data(op_data);
1789 if (S_ISREG(inode->i_mode)) {
1791 if ((attr->ia_valid & ATTR_SIZE) && !hsm_import)
1792 inode_dio_wait(inode);
1793 /* Once we've got the i_mutex, it's safe to set the S_NOSEC
1794 * flag. ll_update_inode (called from ll_md_setattr), clears
1795 * inode flags, so there is a gap where S_NOSEC is not set.
1796 * This can cause a writer to take the i_mutex unnecessarily,
1797 * but this is safe to do and should be rare. */
1798 inode_has_no_xattr(inode);
1801 ll_stats_ops_tally(ll_i2sbi(inode), (attr->ia_valid & ATTR_SIZE) ?
1802 LPROC_LL_TRUNC : LPROC_LL_SETATTR, 1);
1807 int ll_setattr(struct dentry *de, struct iattr *attr)
1809 int mode = de->d_inode->i_mode;
1810 enum op_xvalid xvalid = 0;
1812 if ((attr->ia_valid & (ATTR_CTIME|ATTR_SIZE|ATTR_MODE)) ==
1813 (ATTR_CTIME|ATTR_SIZE|ATTR_MODE))
1814 xvalid |= OP_XVALID_OWNEROVERRIDE;
1816 if (((attr->ia_valid & (ATTR_MODE|ATTR_FORCE|ATTR_SIZE)) ==
1817 (ATTR_SIZE|ATTR_MODE)) &&
1818 (((mode & S_ISUID) && !(attr->ia_mode & S_ISUID)) ||
1819 (((mode & (S_ISGID|S_IXGRP)) == (S_ISGID|S_IXGRP)) &&
1820 !(attr->ia_mode & S_ISGID))))
1821 attr->ia_valid |= ATTR_FORCE;
1823 if ((attr->ia_valid & ATTR_MODE) &&
1825 !(attr->ia_mode & S_ISUID) &&
1826 !(attr->ia_valid & ATTR_KILL_SUID))
1827 attr->ia_valid |= ATTR_KILL_SUID;
1829 if ((attr->ia_valid & ATTR_MODE) &&
1830 ((mode & (S_ISGID|S_IXGRP)) == (S_ISGID|S_IXGRP)) &&
1831 !(attr->ia_mode & S_ISGID) &&
1832 !(attr->ia_valid & ATTR_KILL_SGID))
1833 attr->ia_valid |= ATTR_KILL_SGID;
1835 return ll_setattr_raw(de, attr, xvalid, false);
1838 int ll_statfs_internal(struct ll_sb_info *sbi, struct obd_statfs *osfs,
1841 struct obd_statfs obd_osfs = { 0 };
1846 max_age = ktime_get_seconds() - OBD_STATFS_CACHE_SECONDS;
1848 rc = obd_statfs(NULL, sbi->ll_md_exp, osfs, max_age, flags);
1852 osfs->os_type = LL_SUPER_MAGIC;
1854 CDEBUG(D_SUPER, "MDC blocks %llu/%llu objects %llu/%llu\n",
1855 osfs->os_bavail, osfs->os_blocks, osfs->os_ffree, osfs->os_files);
1857 if (osfs->os_state & OS_STATE_SUM)
1860 if (sbi->ll_flags & LL_SBI_LAZYSTATFS)
1861 flags |= OBD_STATFS_NODELAY;
1863 rc = obd_statfs(NULL, sbi->ll_dt_exp, &obd_osfs, max_age, flags);
1864 if (rc) /* Possibly a filesystem with no OSTs. Report MDT totals. */
1867 CDEBUG(D_SUPER, "OSC blocks %llu/%llu objects %llu/%llu\n",
1868 obd_osfs.os_bavail, obd_osfs.os_blocks, obd_osfs.os_ffree,
1871 osfs->os_bsize = obd_osfs.os_bsize;
1872 osfs->os_blocks = obd_osfs.os_blocks;
1873 osfs->os_bfree = obd_osfs.os_bfree;
1874 osfs->os_bavail = obd_osfs.os_bavail;
1876 /* If we have _some_ OSTs, but don't have as many free objects on the
1877 * OSTs as inodes on the MDTs, reduce the reported number of inodes
1878 * to compensate, so that the "inodes in use" number is correct.
1879 * This should be kept in sync with lod_statfs() behaviour.
1881 if (obd_osfs.os_files && obd_osfs.os_ffree < osfs->os_ffree) {
1882 osfs->os_files = (osfs->os_files - osfs->os_ffree) +
1884 osfs->os_ffree = obd_osfs.os_ffree;
1890 int ll_statfs(struct dentry *de, struct kstatfs *sfs)
1892 struct super_block *sb = de->d_sb;
1893 struct obd_statfs osfs;
1894 __u64 fsid = huge_encode_dev(sb->s_dev);
1897 CDEBUG(D_VFSTRACE, "VFS Op: at %llu jiffies\n", get_jiffies_64());
1898 ll_stats_ops_tally(ll_s2sbi(sb), LPROC_LL_STAFS, 1);
1900 /* Some amount of caching on the client is allowed */
1901 rc = ll_statfs_internal(ll_s2sbi(sb), &osfs, OBD_STATFS_SUM);
1905 statfs_unpack(sfs, &osfs);
1907 /* We need to downshift for all 32-bit kernels, because we can't
1908 * tell if the kernel is being called via sys_statfs64() or not.
1909 * Stop before overflowing f_bsize - in which case it is better
1910 * to just risk EOVERFLOW if caller is using old sys_statfs(). */
1911 if (sizeof(long) < 8) {
1912 while (osfs.os_blocks > ~0UL && sfs->f_bsize < 0x40000000) {
1915 osfs.os_blocks >>= 1;
1916 osfs.os_bfree >>= 1;
1917 osfs.os_bavail >>= 1;
1921 sfs->f_blocks = osfs.os_blocks;
1922 sfs->f_bfree = osfs.os_bfree;
1923 sfs->f_bavail = osfs.os_bavail;
1924 sfs->f_fsid.val[0] = (__u32)fsid;
1925 sfs->f_fsid.val[1] = (__u32)(fsid >> 32);
1929 void ll_inode_size_lock(struct inode *inode)
1931 struct ll_inode_info *lli;
1933 LASSERT(!S_ISDIR(inode->i_mode));
1935 lli = ll_i2info(inode);
1936 mutex_lock(&lli->lli_size_mutex);
1939 void ll_inode_size_unlock(struct inode *inode)
1941 struct ll_inode_info *lli;
1943 lli = ll_i2info(inode);
1944 mutex_unlock(&lli->lli_size_mutex);
1947 void ll_update_inode_flags(struct inode *inode, int ext_flags)
1949 inode->i_flags = ll_ext_to_inode_flags(ext_flags);
1950 if (ext_flags & LUSTRE_PROJINHERIT_FL)
1951 ll_file_set_flag(ll_i2info(inode), LLIF_PROJECT_INHERIT);
1953 ll_file_clear_flag(ll_i2info(inode), LLIF_PROJECT_INHERIT);
1956 int ll_update_inode(struct inode *inode, struct lustre_md *md)
1958 struct ll_inode_info *lli = ll_i2info(inode);
1959 struct mdt_body *body = md->body;
1960 struct ll_sb_info *sbi = ll_i2sbi(inode);
1963 if (body->mbo_valid & OBD_MD_FLEASIZE) {
1964 rc = cl_file_inode_init(inode, md);
1969 if (S_ISDIR(inode->i_mode)) {
1970 rc = ll_update_lsm_md(inode, md);
1975 #ifdef CONFIG_FS_POSIX_ACL
1976 if (body->mbo_valid & OBD_MD_FLACL) {
1977 spin_lock(&lli->lli_lock);
1978 if (lli->lli_posix_acl)
1979 posix_acl_release(lli->lli_posix_acl);
1980 lli->lli_posix_acl = md->posix_acl;
1981 spin_unlock(&lli->lli_lock);
1984 inode->i_ino = cl_fid_build_ino(&body->mbo_fid1,
1985 sbi->ll_flags & LL_SBI_32BIT_API);
1986 inode->i_generation = cl_fid_build_gen(&body->mbo_fid1);
1988 if (body->mbo_valid & OBD_MD_FLATIME) {
1989 if (body->mbo_atime > inode->i_atime.tv_sec)
1990 inode->i_atime.tv_sec = body->mbo_atime;
1991 lli->lli_atime = body->mbo_atime;
1994 if (body->mbo_valid & OBD_MD_FLMTIME) {
1995 if (body->mbo_mtime > inode->i_mtime.tv_sec) {
1997 "setting ino %lu mtime from %lld to %llu\n",
1998 inode->i_ino, (s64)inode->i_mtime.tv_sec,
2000 inode->i_mtime.tv_sec = body->mbo_mtime;
2002 lli->lli_mtime = body->mbo_mtime;
2005 if (body->mbo_valid & OBD_MD_FLCTIME) {
2006 if (body->mbo_ctime > inode->i_ctime.tv_sec)
2007 inode->i_ctime.tv_sec = body->mbo_ctime;
2008 lli->lli_ctime = body->mbo_ctime;
2011 /* Clear i_flags to remove S_NOSEC before permissions are updated */
2012 if (body->mbo_valid & OBD_MD_FLFLAGS)
2013 ll_update_inode_flags(inode, body->mbo_flags);
2014 if (body->mbo_valid & OBD_MD_FLMODE)
2015 inode->i_mode = (inode->i_mode & S_IFMT) |
2016 (body->mbo_mode & ~S_IFMT);
2018 if (body->mbo_valid & OBD_MD_FLTYPE)
2019 inode->i_mode = (inode->i_mode & ~S_IFMT) |
2020 (body->mbo_mode & S_IFMT);
2022 LASSERT(inode->i_mode != 0);
2023 if (S_ISREG(inode->i_mode))
2024 inode->i_blkbits = min(PTLRPC_MAX_BRW_BITS + 1,
2025 LL_MAX_BLKSIZE_BITS);
2027 inode->i_blkbits = inode->i_sb->s_blocksize_bits;
2029 if (body->mbo_valid & OBD_MD_FLUID)
2030 inode->i_uid = make_kuid(&init_user_ns, body->mbo_uid);
2031 if (body->mbo_valid & OBD_MD_FLGID)
2032 inode->i_gid = make_kgid(&init_user_ns, body->mbo_gid);
2033 if (body->mbo_valid & OBD_MD_FLPROJID)
2034 lli->lli_projid = body->mbo_projid;
2035 if (body->mbo_valid & OBD_MD_FLNLINK)
2036 set_nlink(inode, body->mbo_nlink);
2037 if (body->mbo_valid & OBD_MD_FLRDEV)
2038 inode->i_rdev = old_decode_dev(body->mbo_rdev);
2040 if (body->mbo_valid & OBD_MD_FLID) {
2041 /* FID shouldn't be changed! */
2042 if (fid_is_sane(&lli->lli_fid)) {
2043 LASSERTF(lu_fid_eq(&lli->lli_fid, &body->mbo_fid1),
2044 "Trying to change FID "DFID
2045 " to the "DFID", inode "DFID"(%p)\n",
2046 PFID(&lli->lli_fid), PFID(&body->mbo_fid1),
2047 PFID(ll_inode2fid(inode)), inode);
2049 lli->lli_fid = body->mbo_fid1;
2053 LASSERT(fid_seq(&lli->lli_fid) != 0);
2055 if (body->mbo_valid & OBD_MD_FLSIZE) {
2056 i_size_write(inode, body->mbo_size);
2058 CDEBUG(D_VFSTRACE, "inode="DFID", updating i_size %llu\n",
2059 PFID(ll_inode2fid(inode)),
2060 (unsigned long long)body->mbo_size);
2062 if (body->mbo_valid & OBD_MD_FLBLOCKS)
2063 inode->i_blocks = body->mbo_blocks;
2066 if (body->mbo_valid & OBD_MD_TSTATE) {
2067 /* Set LLIF_FILE_RESTORING if restore ongoing and
2068 * clear it when done to ensure to start again
2069 * glimpsing updated attrs
2071 if (body->mbo_t_state & MS_RESTORE)
2072 ll_file_set_flag(lli, LLIF_FILE_RESTORING);
2074 ll_file_clear_flag(lli, LLIF_FILE_RESTORING);
2080 int ll_read_inode2(struct inode *inode, void *opaque)
2082 struct lustre_md *md = opaque;
2083 struct ll_inode_info *lli = ll_i2info(inode);
2087 CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p)\n",
2088 PFID(&lli->lli_fid), inode);
2090 /* Core attributes from the MDS first. This is a new inode, and
2091 * the VFS doesn't zero times in the core inode so we have to do
2092 * it ourselves. They will be overwritten by either MDS or OST
2093 * attributes - we just need to make sure they aren't newer.
2095 inode->i_mtime.tv_sec = 0;
2096 inode->i_atime.tv_sec = 0;
2097 inode->i_ctime.tv_sec = 0;
2099 rc = ll_update_inode(inode, md);
2103 /* OIDEBUG(inode); */
2105 #ifdef HAVE_BACKING_DEV_INFO
2106 /* initializing backing dev info. */
2107 inode->i_mapping->backing_dev_info = &s2lsi(inode->i_sb)->lsi_bdi;
2109 if (S_ISREG(inode->i_mode)) {
2110 struct ll_sb_info *sbi = ll_i2sbi(inode);
2111 inode->i_op = &ll_file_inode_operations;
2112 inode->i_fop = sbi->ll_fop;
2113 inode->i_mapping->a_ops = (struct address_space_operations *)&ll_aops;
2115 } else if (S_ISDIR(inode->i_mode)) {
2116 inode->i_op = &ll_dir_inode_operations;
2117 inode->i_fop = &ll_dir_operations;
2119 } else if (S_ISLNK(inode->i_mode)) {
2120 inode->i_op = &ll_fast_symlink_inode_operations;
2123 inode->i_op = &ll_special_inode_operations;
2125 init_special_inode(inode, inode->i_mode,
2134 void ll_delete_inode(struct inode *inode)
2136 struct ll_inode_info *lli = ll_i2info(inode);
2137 struct address_space *mapping = &inode->i_data;
2138 unsigned long nrpages;
2141 if (S_ISREG(inode->i_mode) && lli->lli_clob != NULL) {
2142 /* It is last chance to write out dirty pages,
2143 * otherwise we may lose data while umount.
2145 * If i_nlink is 0 then just discard data. This is safe because
2146 * local inode gets i_nlink 0 from server only for the last
2147 * unlink, so that file is not opened somewhere else
2149 cl_sync_file_range(inode, 0, OBD_OBJECT_EOF, inode->i_nlink ?
2150 CL_FSYNC_LOCAL : CL_FSYNC_DISCARD, 1);
2152 truncate_inode_pages_final(mapping);
2154 /* Workaround for LU-118: Note nrpages may not be totally updated when
2155 * truncate_inode_pages() returns, as there can be a page in the process
2156 * of deletion (inside __delete_from_page_cache()) in the specified
2157 * range. Thus mapping->nrpages can be non-zero when this function
2158 * returns even after truncation of the whole mapping. Only do this if
2159 * npages isn't already zero.
2161 nrpages = mapping->nrpages;
2163 xa_lock_irq(&mapping->i_pages);
2164 nrpages = mapping->nrpages;
2165 xa_unlock_irq(&mapping->i_pages);
2166 } /* Workaround end */
2168 LASSERTF(nrpages == 0, "%s: inode="DFID"(%p) nrpages=%lu, "
2169 "see https://jira.whamcloud.com/browse/LU-118\n",
2170 ll_i2sbi(inode)->ll_fsname,
2171 PFID(ll_inode2fid(inode)), inode, nrpages);
2173 #ifdef HAVE_SBOPS_EVICT_INODE
2174 ll_clear_inode(inode);
2181 int ll_iocontrol(struct inode *inode, struct file *file,
2182 unsigned int cmd, unsigned long arg)
2184 struct ll_sb_info *sbi = ll_i2sbi(inode);
2185 struct ptlrpc_request *req = NULL;
2190 case FS_IOC_GETFLAGS: {
2191 struct mdt_body *body;
2192 struct md_op_data *op_data;
2194 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL,
2195 0, 0, LUSTRE_OPC_ANY,
2197 if (IS_ERR(op_data))
2198 RETURN(PTR_ERR(op_data));
2200 op_data->op_valid = OBD_MD_FLFLAGS;
2201 rc = md_getattr(sbi->ll_md_exp, op_data, &req);
2202 ll_finish_md_op_data(op_data);
2204 CERROR("%s: failure inode "DFID": rc = %d\n",
2205 sbi->ll_md_exp->exp_obd->obd_name,
2206 PFID(ll_inode2fid(inode)), rc);
2210 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
2212 flags = body->mbo_flags;
2214 ptlrpc_req_finished(req);
2216 RETURN(put_user(flags, (int __user *)arg));
2218 case FS_IOC_SETFLAGS: {
2220 struct md_op_data *op_data;
2221 struct cl_object *obj;
2222 struct fsxattr fa = { 0 };
2224 if (get_user(flags, (int __user *)arg))
2227 fa.fsx_projid = ll_i2info(inode)->lli_projid;
2228 if (flags & LUSTRE_PROJINHERIT_FL)
2229 fa.fsx_xflags = FS_XFLAG_PROJINHERIT;
2231 rc = ll_ioctl_check_project(inode, &fa);
2235 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
2236 LUSTRE_OPC_ANY, NULL);
2237 if (IS_ERR(op_data))
2238 RETURN(PTR_ERR(op_data));
2240 op_data->op_attr_flags = flags;
2241 op_data->op_xvalid |= OP_XVALID_FLAGS;
2242 rc = md_setattr(sbi->ll_md_exp, op_data, NULL, 0, &req);
2243 ll_finish_md_op_data(op_data);
2244 ptlrpc_req_finished(req);
2248 ll_update_inode_flags(inode, flags);
2250 obj = ll_i2info(inode)->lli_clob;
2254 OBD_ALLOC_PTR(attr);
2258 rc = cl_setattr_ost(obj, attr, OP_XVALID_FLAGS, flags);
2270 int ll_flush_ctx(struct inode *inode)
2272 struct ll_sb_info *sbi = ll_i2sbi(inode);
2274 CDEBUG(D_SEC, "flush context for user %d\n",
2275 from_kuid(&init_user_ns, current_uid()));
2277 obd_set_info_async(NULL, sbi->ll_md_exp,
2278 sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
2280 obd_set_info_async(NULL, sbi->ll_dt_exp,
2281 sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
2286 /* umount -f client means force down, don't save state */
2287 void ll_umount_begin(struct super_block *sb)
2289 struct ll_sb_info *sbi = ll_s2sbi(sb);
2290 struct obd_device *obd;
2291 struct obd_ioctl_data *ioc_data;
2292 struct l_wait_info lwi;
2293 wait_queue_head_t waitq;
2296 CDEBUG(D_VFSTRACE, "VFS Op: superblock %p count %d active %d\n", sb,
2297 sb->s_count, atomic_read(&sb->s_active));
2299 obd = class_exp2obd(sbi->ll_md_exp);
2301 CERROR("Invalid MDC connection handle %#llx\n",
2302 sbi->ll_md_exp->exp_handle.h_cookie);
2308 obd = class_exp2obd(sbi->ll_dt_exp);
2310 CERROR("Invalid LOV connection handle %#llx\n",
2311 sbi->ll_dt_exp->exp_handle.h_cookie);
2317 OBD_ALLOC_PTR(ioc_data);
2319 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_md_exp,
2320 sizeof *ioc_data, ioc_data, NULL);
2322 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_dt_exp,
2323 sizeof *ioc_data, ioc_data, NULL);
2325 OBD_FREE_PTR(ioc_data);
2328 /* Really, we'd like to wait until there are no requests outstanding,
2329 * and then continue. For now, we just periodically checking for vfs
2330 * to decrement mnt_cnt and hope to finish it within 10sec.
2332 init_waitqueue_head(&waitq);
2333 lwi = LWI_TIMEOUT_INTERVAL(cfs_time_seconds(10),
2334 cfs_time_seconds(1), NULL, NULL);
2335 l_wait_event(waitq, may_umount(sbi->ll_mnt.mnt), &lwi);
2340 int ll_remount_fs(struct super_block *sb, int *flags, char *data)
2342 struct ll_sb_info *sbi = ll_s2sbi(sb);
2343 char *profilenm = get_profile_name(sb);
2347 if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY)) {
2348 read_only = *flags & MS_RDONLY;
2349 err = obd_set_info_async(NULL, sbi->ll_md_exp,
2350 sizeof(KEY_READ_ONLY),
2351 KEY_READ_ONLY, sizeof(read_only),
2354 LCONSOLE_WARN("Failed to remount %s %s (%d)\n",
2355 profilenm, read_only ?
2356 "read-only" : "read-write", err);
2361 sb->s_flags |= MS_RDONLY;
2363 sb->s_flags &= ~MS_RDONLY;
2365 if (sbi->ll_flags & LL_SBI_VERBOSE)
2366 LCONSOLE_WARN("Remounted %s %s\n", profilenm,
2367 read_only ? "read-only" : "read-write");
2373 * Cleanup the open handle that is cached on MDT-side.
2375 * For open case, the client side open handling thread may hit error
2376 * after the MDT grant the open. Under such case, the client should
2377 * send close RPC to the MDT as cleanup; otherwise, the open handle
2378 * on the MDT will be leaked there until the client umount or evicted.
2380 * In further, if someone unlinked the file, because the open handle
2381 * holds the reference on such file/object, then it will block the
2382 * subsequent threads that want to locate such object via FID.
2384 * \param[in] sb super block for this file-system
2385 * \param[in] open_req pointer to the original open request
2387 void ll_open_cleanup(struct super_block *sb, struct ptlrpc_request *open_req)
2389 struct mdt_body *body;
2390 struct md_op_data *op_data;
2391 struct ptlrpc_request *close_req = NULL;
2392 struct obd_export *exp = ll_s2sbi(sb)->ll_md_exp;
2395 body = req_capsule_server_get(&open_req->rq_pill, &RMF_MDT_BODY);
2396 OBD_ALLOC_PTR(op_data);
2397 if (op_data == NULL) {
2398 CWARN("%s: cannot allocate op_data to release open handle for "
2399 DFID"\n", ll_s2sbi(sb)->ll_fsname, PFID(&body->mbo_fid1));
2404 op_data->op_fid1 = body->mbo_fid1;
2405 op_data->op_open_handle = body->mbo_open_handle;
2406 op_data->op_mod_time = ktime_get_real_seconds();
2407 md_close(exp, op_data, NULL, &close_req);
2408 ptlrpc_req_finished(close_req);
2409 ll_finish_md_op_data(op_data);
2414 int ll_prep_inode(struct inode **inode, struct ptlrpc_request *req,
2415 struct super_block *sb, struct lookup_intent *it)
2417 struct ll_sb_info *sbi = NULL;
2418 struct lustre_md md = { NULL };
2422 LASSERT(*inode || sb);
2423 sbi = sb ? ll_s2sbi(sb) : ll_i2sbi(*inode);
2424 rc = md_get_lustre_md(sbi->ll_md_exp, req, sbi->ll_dt_exp,
2425 sbi->ll_md_exp, &md);
2430 rc = ll_update_inode(*inode, &md);
2434 LASSERT(sb != NULL);
2437 * At this point server returns to client's same fid as client
2438 * generated for creating. So using ->fid1 is okay here.
2440 if (!fid_is_sane(&md.body->mbo_fid1)) {
2441 CERROR("%s: Fid is insane "DFID"\n",
2443 PFID(&md.body->mbo_fid1));
2444 GOTO(out, rc = -EINVAL);
2447 *inode = ll_iget(sb, cl_fid_build_ino(&md.body->mbo_fid1,
2448 sbi->ll_flags & LL_SBI_32BIT_API),
2450 if (IS_ERR(*inode)) {
2451 #ifdef CONFIG_FS_POSIX_ACL
2453 posix_acl_release(md.posix_acl);
2454 md.posix_acl = NULL;
2457 rc = IS_ERR(*inode) ? PTR_ERR(*inode) : -ENOMEM;
2459 CERROR("new_inode -fatal: rc %d\n", rc);
2464 /* Handling piggyback layout lock.
2465 * Layout lock can be piggybacked by getattr and open request.
2466 * The lsm can be applied to inode only if it comes with a layout lock
2467 * otherwise correct layout may be overwritten, for example:
2468 * 1. proc1: mdt returns a lsm but not granting layout
2469 * 2. layout was changed by another client
2470 * 3. proc2: refresh layout and layout lock granted
2471 * 4. proc1: to apply a stale layout */
2472 if (it != NULL && it->it_lock_mode != 0) {
2473 struct lustre_handle lockh;
2474 struct ldlm_lock *lock;
2476 lockh.cookie = it->it_lock_handle;
2477 lock = ldlm_handle2lock(&lockh);
2478 LASSERT(lock != NULL);
2479 if (ldlm_has_layout(lock)) {
2480 struct cl_object_conf conf;
2482 memset(&conf, 0, sizeof(conf));
2483 conf.coc_opc = OBJECT_CONF_SET;
2484 conf.coc_inode = *inode;
2485 conf.coc_lock = lock;
2486 conf.u.coc_layout = md.layout;
2487 (void)ll_layout_conf(*inode, &conf);
2489 LDLM_LOCK_PUT(lock);
2495 /* cleanup will be done if necessary */
2496 md_free_lustre_md(sbi->ll_md_exp, &md);
2498 if (rc != 0 && it != NULL && it->it_op & IT_OPEN)
2499 ll_open_cleanup(sb != NULL ? sb : (*inode)->i_sb, req);
2504 int ll_obd_statfs(struct inode *inode, void __user *arg)
2506 struct ll_sb_info *sbi = NULL;
2507 struct obd_export *exp;
2509 struct obd_ioctl_data *data = NULL;
2513 if (!inode || !(sbi = ll_i2sbi(inode)))
2514 GOTO(out_statfs, rc = -EINVAL);
2516 rc = obd_ioctl_getdata(&buf, &len, arg);
2518 GOTO(out_statfs, rc);
2521 if (!data->ioc_inlbuf1 || !data->ioc_inlbuf2 ||
2522 !data->ioc_pbuf1 || !data->ioc_pbuf2)
2523 GOTO(out_statfs, rc = -EINVAL);
2525 if (data->ioc_inllen1 != sizeof(__u32) ||
2526 data->ioc_inllen2 != sizeof(__u32) ||
2527 data->ioc_plen1 != sizeof(struct obd_statfs) ||
2528 data->ioc_plen2 != sizeof(struct obd_uuid))
2529 GOTO(out_statfs, rc = -EINVAL);
2531 memcpy(&type, data->ioc_inlbuf1, sizeof(__u32));
2532 if (type & LL_STATFS_LMV)
2533 exp = sbi->ll_md_exp;
2534 else if (type & LL_STATFS_LOV)
2535 exp = sbi->ll_dt_exp;
2537 GOTO(out_statfs, rc = -ENODEV);
2539 rc = obd_iocontrol(IOC_OBD_STATFS, exp, len, buf, NULL);
2541 GOTO(out_statfs, rc);
2543 OBD_FREE_LARGE(buf, len);
2548 * this is normally called in ll_fini_md_op_data(), but sometimes it needs to
2549 * be called early to avoid deadlock.
2551 void ll_unlock_md_op_lsm(struct md_op_data *op_data)
2553 if (op_data->op_mea2_sem) {
2554 up_read(op_data->op_mea2_sem);
2555 op_data->op_mea2_sem = NULL;
2558 if (op_data->op_mea1_sem) {
2559 up_read(op_data->op_mea1_sem);
2560 op_data->op_mea1_sem = NULL;
2564 /* this function prepares md_op_data hint for passing it down to MD stack. */
2565 struct md_op_data *ll_prep_md_op_data(struct md_op_data *op_data,
2566 struct inode *i1, struct inode *i2,
2567 const char *name, size_t namelen,
2568 __u32 mode, __u32 opc, void *data)
2570 LASSERT(i1 != NULL);
2573 /* Do not reuse namelen for something else. */
2575 return ERR_PTR(-EINVAL);
2577 if (namelen > ll_i2sbi(i1)->ll_namelen)
2578 return ERR_PTR(-ENAMETOOLONG);
2580 if (!lu_name_is_valid_2(name, namelen))
2581 return ERR_PTR(-EINVAL);
2584 if (op_data == NULL)
2585 OBD_ALLOC_PTR(op_data);
2587 if (op_data == NULL)
2588 return ERR_PTR(-ENOMEM);
2590 ll_i2gids(op_data->op_suppgids, i1, i2);
2591 op_data->op_fid1 = *ll_inode2fid(i1);
2592 op_data->op_default_stripe_offset = -1;
2594 if (S_ISDIR(i1->i_mode)) {
2595 down_read(&ll_i2info(i1)->lli_lsm_sem);
2596 op_data->op_mea1_sem = &ll_i2info(i1)->lli_lsm_sem;
2597 op_data->op_mea1 = ll_i2info(i1)->lli_lsm_md;
2598 if (opc == LUSTRE_OPC_MKDIR)
2599 op_data->op_default_stripe_offset =
2600 ll_i2info(i1)->lli_def_stripe_offset;
2604 op_data->op_fid2 = *ll_inode2fid(i2);
2605 if (S_ISDIR(i2->i_mode)) {
2607 down_read(&ll_i2info(i2)->lli_lsm_sem);
2608 op_data->op_mea2_sem =
2609 &ll_i2info(i2)->lli_lsm_sem;
2611 op_data->op_mea2 = ll_i2info(i2)->lli_lsm_md;
2614 fid_zero(&op_data->op_fid2);
2617 if (ll_i2sbi(i1)->ll_flags & LL_SBI_64BIT_HASH)
2618 op_data->op_cli_flags |= CLI_HASH64;
2620 if (ll_need_32bit_api(ll_i2sbi(i1)))
2621 op_data->op_cli_flags |= CLI_API32;
2623 op_data->op_name = name;
2624 op_data->op_namelen = namelen;
2625 op_data->op_mode = mode;
2626 op_data->op_mod_time = ktime_get_real_seconds();
2627 op_data->op_fsuid = from_kuid(&init_user_ns, current_fsuid());
2628 op_data->op_fsgid = from_kgid(&init_user_ns, current_fsgid());
2629 op_data->op_cap = cfs_curproc_cap_pack();
2630 op_data->op_mds = 0;
2631 if ((opc == LUSTRE_OPC_CREATE) && (name != NULL) &&
2632 filename_is_volatile(name, namelen, &op_data->op_mds)) {
2633 op_data->op_bias |= MDS_CREATE_VOLATILE;
2635 op_data->op_data = data;
2640 void ll_finish_md_op_data(struct md_op_data *op_data)
2642 ll_unlock_md_op_lsm(op_data);
2643 security_release_secctx(op_data->op_file_secctx,
2644 op_data->op_file_secctx_size);
2645 OBD_FREE_PTR(op_data);
2648 #ifdef HAVE_SUPEROPS_USE_DENTRY
2649 int ll_show_options(struct seq_file *seq, struct dentry *dentry)
2651 int ll_show_options(struct seq_file *seq, struct vfsmount *vfs)
2654 struct ll_sb_info *sbi;
2656 #ifdef HAVE_SUPEROPS_USE_DENTRY
2657 LASSERT((seq != NULL) && (dentry != NULL));
2658 sbi = ll_s2sbi(dentry->d_sb);
2660 LASSERT((seq != NULL) && (vfs != NULL));
2661 sbi = ll_s2sbi(vfs->mnt_sb);
2664 if (sbi->ll_flags & LL_SBI_NOLCK)
2665 seq_puts(seq, ",nolock");
2667 if (sbi->ll_flags & LL_SBI_FLOCK)
2668 seq_puts(seq, ",flock");
2670 if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
2671 seq_puts(seq, ",localflock");
2673 if (sbi->ll_flags & LL_SBI_USER_XATTR)
2674 seq_puts(seq, ",user_xattr");
2676 if (sbi->ll_flags & LL_SBI_LAZYSTATFS)
2677 seq_puts(seq, ",lazystatfs");
2679 if (sbi->ll_flags & LL_SBI_USER_FID2PATH)
2680 seq_puts(seq, ",user_fid2path");
2682 if (sbi->ll_flags & LL_SBI_ALWAYS_PING)
2683 seq_puts(seq, ",always_ping");
2689 * Get obd name by cmd, and copy out to user space
2691 int ll_get_obd_name(struct inode *inode, unsigned int cmd, unsigned long arg)
2693 struct ll_sb_info *sbi = ll_i2sbi(inode);
2694 struct obd_device *obd;
2697 if (cmd == OBD_IOC_GETDTNAME)
2698 obd = class_exp2obd(sbi->ll_dt_exp);
2699 else if (cmd == OBD_IOC_GETMDNAME)
2700 obd = class_exp2obd(sbi->ll_md_exp);
2707 if (copy_to_user((void __user *)arg, obd->obd_name,
2708 strlen(obd->obd_name) + 1))
2714 static char* ll_d_path(struct dentry *dentry, char *buf, int bufsize)
2721 p.mnt = current->fs->root.mnt;
2723 path = d_path(&p, buf, bufsize);
2728 void ll_dirty_page_discard_warn(struct page *page, int ioret)
2730 char *buf, *path = NULL;
2731 struct dentry *dentry = NULL;
2732 struct inode *inode = page->mapping->host;
2734 /* this can be called inside spin lock so use GFP_ATOMIC. */
2735 buf = (char *)__get_free_page(GFP_ATOMIC);
2737 dentry = d_find_alias(page->mapping->host);
2739 path = ll_d_path(dentry, buf, PAGE_SIZE);
2743 "%s: dirty page discard: %s/fid: "DFID"/%s may get corrupted "
2744 "(rc %d)\n", ll_i2sbi(inode)->ll_fsname,
2745 s2lsi(page->mapping->host->i_sb)->lsi_lmd->lmd_dev,
2746 PFID(ll_inode2fid(inode)),
2747 (path && !IS_ERR(path)) ? path : "", ioret);
2753 free_page((unsigned long)buf);
2756 ssize_t ll_copy_user_md(const struct lov_user_md __user *md,
2757 struct lov_user_md **kbuf)
2759 struct lov_user_md lum;
2763 if (copy_from_user(&lum, md, sizeof(lum)))
2766 lum_size = ll_lov_user_md_size(&lum);
2770 OBD_ALLOC(*kbuf, lum_size);
2774 if (copy_from_user(*kbuf, md, lum_size) != 0) {
2775 OBD_FREE(*kbuf, lum_size);
2783 * Compute llite root squash state after a change of root squash
2784 * configuration setting or add/remove of a lnet nid
2786 void ll_compute_rootsquash_state(struct ll_sb_info *sbi)
2788 struct root_squash_info *squash = &sbi->ll_squash;
2791 struct lnet_process_id id;
2793 /* Update norootsquash flag */
2794 down_write(&squash->rsi_sem);
2795 if (list_empty(&squash->rsi_nosquash_nids))
2796 sbi->ll_flags &= ~LL_SBI_NOROOTSQUASH;
2798 /* Do not apply root squash as soon as one of our NIDs is
2799 * in the nosquash_nids list */
2802 while (LNetGetId(i++, &id) != -ENOENT) {
2803 if (LNET_NETTYP(LNET_NIDNET(id.nid)) == LOLND)
2805 if (cfs_match_nid(id.nid, &squash->rsi_nosquash_nids)) {
2811 sbi->ll_flags |= LL_SBI_NOROOTSQUASH;
2813 sbi->ll_flags &= ~LL_SBI_NOROOTSQUASH;
2815 up_write(&squash->rsi_sem);
2819 * Parse linkea content to extract information about a given hardlink
2821 * \param[in] ldata - Initialized linkea data
2822 * \param[in] linkno - Link identifier
2823 * \param[out] parent_fid - The entry's parent FID
2824 * \param[out] ln - Entry name destination buffer
2826 * \retval 0 on success
2827 * \retval Appropriate negative error code on failure
2829 static int ll_linkea_decode(struct linkea_data *ldata, unsigned int linkno,
2830 struct lu_fid *parent_fid, struct lu_name *ln)
2836 rc = linkea_init_with_rec(ldata);
2840 if (linkno >= ldata->ld_leh->leh_reccount)
2841 /* beyond last link */
2844 linkea_first_entry(ldata);
2845 for (idx = 0; ldata->ld_lee != NULL; idx++) {
2846 linkea_entry_unpack(ldata->ld_lee, &ldata->ld_reclen, ln,
2851 linkea_next_entry(ldata);
2861 * Get parent FID and name of an identified link. Operation is performed for
2862 * a given link number, letting the caller iterate over linkno to list one or
2863 * all links of an entry.
2865 * \param[in] file - File descriptor against which to perform the operation
2866 * \param[in,out] arg - User-filled structure containing the linkno to operate
2867 * on and the available size. It is eventually filled with
2868 * the requested information or left untouched on error
2870 * \retval - 0 on success
2871 * \retval - Appropriate negative error code on failure
2873 int ll_getparent(struct file *file, struct getparent __user *arg)
2875 struct inode *inode = file_inode(file);
2876 struct linkea_data *ldata;
2877 struct lu_buf buf = LU_BUF_NULL;
2879 struct lu_fid parent_fid;
2886 if (!cfs_capable(CFS_CAP_DAC_READ_SEARCH) &&
2887 !(ll_i2sbi(inode)->ll_flags & LL_SBI_USER_FID2PATH))
2890 if (get_user(name_size, &arg->gp_name_size))
2893 if (get_user(linkno, &arg->gp_linkno))
2896 if (name_size > PATH_MAX)
2899 OBD_ALLOC(ldata, sizeof(*ldata));
2903 rc = linkea_data_new(ldata, &buf);
2905 GOTO(ldata_free, rc);
2907 #ifdef HAVE_XATTR_HANDLER_FLAGS
2908 rc = ll_xattr_list(inode, XATTR_NAME_LINK, XATTR_TRUSTED_T, buf.lb_buf,
2909 buf.lb_len, OBD_MD_FLXATTR);
2911 rc = ll_getxattr(file_dentry(file), XATTR_NAME_LINK, buf.lb_buf,
2913 #endif /* HAVE_XATTR_HANDLER_FLAGS */
2917 rc = ll_linkea_decode(ldata, linkno, &parent_fid, &ln);
2921 if (ln.ln_namelen >= name_size)
2922 GOTO(lb_free, rc = -EOVERFLOW);
2924 if (copy_to_user(&arg->gp_fid, &parent_fid, sizeof(arg->gp_fid)))
2925 GOTO(lb_free, rc = -EFAULT);
2927 if (copy_to_user(&arg->gp_name, ln.ln_name, ln.ln_namelen))
2928 GOTO(lb_free, rc = -EFAULT);
2930 if (put_user('\0', arg->gp_name + ln.ln_namelen))
2931 GOTO(lb_free, rc = -EFAULT);
2936 OBD_FREE(ldata, sizeof(*ldata));