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 #include <lustre_ha.h>
53 #include <lustre_dlm.h>
54 #include <lprocfs_status.h>
55 #include <lustre_disk.h>
56 #include <uapi/linux/lustre/lustre_param.h>
57 #include <lustre_log.h>
58 #include <cl_object.h>
59 #include <obd_cksum.h>
60 #include "llite_internal.h"
62 struct kmem_cache *ll_file_data_slab;
65 #define log2(n) ffz(~(n))
68 static struct ll_sb_info *ll_init_sbi(void)
70 struct ll_sb_info *sbi = NULL;
72 unsigned long lru_page_max;
82 spin_lock_init(&sbi->ll_lock);
83 mutex_init(&sbi->ll_lco.lco_lock);
84 spin_lock_init(&sbi->ll_pp_extent_lock);
85 spin_lock_init(&sbi->ll_process_lock);
86 sbi->ll_rw_stats_on = 0;
89 pages = si.totalram - si.totalhigh;
90 lru_page_max = pages / 2;
92 /* initialize ll_cache data */
93 sbi->ll_cache = cl_cache_init(lru_page_max);
94 if (sbi->ll_cache == NULL) {
95 OBD_FREE(sbi, sizeof(*sbi));
99 sbi->ll_ra_info.ra_max_pages_per_file = min(pages / 32,
100 SBI_DEFAULT_READAHEAD_MAX);
101 sbi->ll_ra_info.ra_max_pages = sbi->ll_ra_info.ra_max_pages_per_file;
102 sbi->ll_ra_info.ra_max_read_ahead_whole_pages = -1;
104 ll_generate_random_uuid(uuid);
105 class_uuid_unparse(uuid, &sbi->ll_sb_uuid);
106 CDEBUG(D_CONFIG, "generated uuid: %s\n", sbi->ll_sb_uuid.uuid);
108 sbi->ll_flags |= LL_SBI_VERBOSE;
109 #ifdef ENABLE_CHECKSUM
110 sbi->ll_flags |= LL_SBI_CHECKSUM;
113 sbi->ll_flags |= LL_SBI_FLOCK;
116 #ifdef HAVE_LRU_RESIZE_SUPPORT
117 sbi->ll_flags |= LL_SBI_LRU_RESIZE;
119 sbi->ll_flags |= LL_SBI_LAZYSTATFS;
121 for (i = 0; i <= LL_PROCESS_HIST_MAX; i++) {
122 spin_lock_init(&sbi->ll_rw_extents_info.pp_extents[i].
124 spin_lock_init(&sbi->ll_rw_extents_info.pp_extents[i].
128 /* metadata statahead is enabled by default */
129 sbi->ll_sa_running_max = LL_SA_RUNNING_DEF;
130 sbi->ll_sa_max = LL_SA_RPC_DEF;
131 atomic_set(&sbi->ll_sa_total, 0);
132 atomic_set(&sbi->ll_sa_wrong, 0);
133 atomic_set(&sbi->ll_sa_running, 0);
134 atomic_set(&sbi->ll_agl_total, 0);
135 sbi->ll_flags |= LL_SBI_AGL_ENABLED;
136 sbi->ll_flags |= LL_SBI_FAST_READ;
137 sbi->ll_flags |= LL_SBI_TINY_WRITE;
140 sbi->ll_squash.rsi_uid = 0;
141 sbi->ll_squash.rsi_gid = 0;
142 INIT_LIST_HEAD(&sbi->ll_squash.rsi_nosquash_nids);
143 init_rwsem(&sbi->ll_squash.rsi_sem);
145 /* Per-filesystem file heat */
146 sbi->ll_heat_decay_weight = SBI_DEFAULT_HEAT_DECAY_WEIGHT;
147 sbi->ll_heat_period_second = SBI_DEFAULT_HEAT_PERIOD_SECOND;
151 static void ll_free_sbi(struct super_block *sb)
153 struct ll_sb_info *sbi = ll_s2sbi(sb);
157 if (!list_empty(&sbi->ll_squash.rsi_nosquash_nids))
158 cfs_free_nidlist(&sbi->ll_squash.rsi_nosquash_nids);
159 if (sbi->ll_cache != NULL) {
160 cl_cache_decref(sbi->ll_cache);
161 sbi->ll_cache = NULL;
163 OBD_FREE(sbi, sizeof(*sbi));
168 static int client_common_fill_super(struct super_block *sb, char *md, char *dt,
169 struct vfsmount *mnt)
171 struct inode *root = NULL;
172 struct ll_sb_info *sbi = ll_s2sbi(sb);
173 struct obd_statfs *osfs = NULL;
174 struct ptlrpc_request *request = NULL;
175 struct obd_connect_data *data = NULL;
176 struct obd_uuid *uuid;
177 struct md_op_data *op_data;
178 struct lustre_md lmd;
180 int size, err, checksum;
183 sbi->ll_md_obd = class_name2obd(md);
184 if (!sbi->ll_md_obd) {
185 CERROR("MD %s: not setup or attached\n", md);
199 /* pass client page size via ocd_grant_blkbits, the server should report
200 * back its backend blocksize for grant calculation purpose */
201 data->ocd_grant_blkbits = PAGE_SHIFT;
203 /* indicate MDT features supported by this client */
204 data->ocd_connect_flags = OBD_CONNECT_IBITS | OBD_CONNECT_NODEVOH |
205 OBD_CONNECT_ATTRFID | OBD_CONNECT_GRANT |
206 OBD_CONNECT_VERSION | OBD_CONNECT_BRW_SIZE |
207 OBD_CONNECT_SRVLOCK | OBD_CONNECT_TRUNCLOCK|
208 OBD_CONNECT_MDS_CAPA | OBD_CONNECT_OSS_CAPA |
209 OBD_CONNECT_CANCELSET | OBD_CONNECT_FID |
210 OBD_CONNECT_AT | OBD_CONNECT_LOV_V3 |
211 OBD_CONNECT_VBR | OBD_CONNECT_FULL20 |
212 OBD_CONNECT_64BITHASH |
213 OBD_CONNECT_EINPROGRESS |
214 OBD_CONNECT_JOBSTATS | OBD_CONNECT_LVB_TYPE |
215 OBD_CONNECT_LAYOUTLOCK | OBD_CONNECT_PINGLESS|
216 OBD_CONNECT_MAX_EASIZE |
217 OBD_CONNECT_FLOCK_DEAD |
218 OBD_CONNECT_DISP_STRIPE | OBD_CONNECT_LFSCK |
219 OBD_CONNECT_OPEN_BY_FID |
220 OBD_CONNECT_DIR_STRIPE |
221 OBD_CONNECT_BULK_MBITS | OBD_CONNECT_CKSUM |
222 OBD_CONNECT_SUBTREE |
223 OBD_CONNECT_MULTIMODRPCS |
224 OBD_CONNECT_GRANT_PARAM |
225 OBD_CONNECT_SHORTIO | OBD_CONNECT_FLAGS2;
227 data->ocd_connect_flags2 = OBD_CONNECT2_FLR |
228 OBD_CONNECT2_LOCK_CONVERT |
229 OBD_CONNECT2_DIR_MIGRATE |
230 OBD_CONNECT2_SUM_STATFS |
231 OBD_CONNECT2_ARCHIVE_ID_ARRAY;
233 #ifdef HAVE_LRU_RESIZE_SUPPORT
234 if (sbi->ll_flags & LL_SBI_LRU_RESIZE)
235 data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
237 #ifdef CONFIG_FS_POSIX_ACL
238 data->ocd_connect_flags |= OBD_CONNECT_ACL | OBD_CONNECT_UMASK |
239 OBD_CONNECT_LARGE_ACL;
242 data->ocd_cksum_types = obd_cksum_types_supported_client();
244 if (OBD_FAIL_CHECK(OBD_FAIL_MDC_LIGHTWEIGHT))
245 /* flag mdc connection as lightweight, only used for test
246 * purpose, use with care */
247 data->ocd_connect_flags |= OBD_CONNECT_LIGHTWEIGHT;
249 data->ocd_ibits_known = MDS_INODELOCK_FULL;
250 data->ocd_version = LUSTRE_VERSION_CODE;
252 if (sb->s_flags & MS_RDONLY)
253 data->ocd_connect_flags |= OBD_CONNECT_RDONLY;
254 if (sbi->ll_flags & LL_SBI_USER_XATTR)
255 data->ocd_connect_flags |= OBD_CONNECT_XATTR;
258 /* Setting this indicates we correctly support S_NOSEC (See kernel
259 * commit 9e1f1de02c2275d7172e18dc4e7c2065777611bf)
261 sb->s_flags |= MS_NOSEC;
264 if (sbi->ll_flags & LL_SBI_FLOCK)
265 sbi->ll_fop = &ll_file_operations_flock;
266 else if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
267 sbi->ll_fop = &ll_file_operations;
269 sbi->ll_fop = &ll_file_operations_noflock;
271 /* always ping even if server suppress_pings */
272 if (sbi->ll_flags & LL_SBI_ALWAYS_PING)
273 data->ocd_connect_flags &= ~OBD_CONNECT_PINGLESS;
275 obd_connect_set_secctx(data);
277 #if defined(CONFIG_SECURITY)
278 data->ocd_connect_flags2 |= OBD_CONNECT2_SELINUX_POLICY;
281 data->ocd_brw_size = MD_MAX_BRW_SIZE;
283 err = obd_connect(NULL, &sbi->ll_md_exp, sbi->ll_md_obd,
284 &sbi->ll_sb_uuid, data, sbi->ll_cache);
286 LCONSOLE_ERROR_MSG(0x14f, "An MDT (md %s) is performing "
287 "recovery, of which this client is not a "
288 "part. Please wait for recovery to complete,"
289 " abort, or time out.\n", md);
292 CERROR("cannot connect to %s: rc = %d\n", md, err);
296 sbi->ll_md_exp->exp_connect_data = *data;
298 err = obd_fid_init(sbi->ll_md_exp->exp_obd, sbi->ll_md_exp,
299 LUSTRE_SEQ_METADATA);
301 CERROR("%s: Can't init metadata layer FID infrastructure, "
302 "rc = %d\n", sbi->ll_md_exp->exp_obd->obd_name, err);
306 /* For mount, we only need fs info from MDT0, and also in DNE, it
307 * can make sure the client can be mounted as long as MDT0 is
309 err = obd_statfs(NULL, sbi->ll_md_exp, osfs,
310 ktime_get_seconds() - OBD_STATFS_CACHE_SECONDS,
311 OBD_STATFS_FOR_MDT0);
313 GOTO(out_md_fid, err);
315 /* This needs to be after statfs to ensure connect has finished.
316 * Note that "data" does NOT contain the valid connect reply.
317 * If connecting to a 1.8 server there will be no LMV device, so
318 * we can access the MDC export directly and exp_connect_flags will
319 * be non-zero, but if accessing an upgraded 2.1 server it will
320 * have the correct flags filled in.
321 * XXX: fill in the LMV exp_connect_flags from MDC(s). */
322 valid = exp_connect_flags(sbi->ll_md_exp) & CLIENT_CONNECT_MDT_REQD;
323 if (exp_connect_flags(sbi->ll_md_exp) != 0 &&
324 valid != CLIENT_CONNECT_MDT_REQD) {
327 OBD_ALLOC_WAIT(buf, PAGE_SIZE);
328 obd_connect_flags2str(buf, PAGE_SIZE,
329 valid ^ CLIENT_CONNECT_MDT_REQD, 0, ",");
330 LCONSOLE_ERROR_MSG(0x170, "Server %s does not support "
331 "feature(s) needed for correct operation "
332 "of this client (%s). Please upgrade "
333 "server or downgrade client.\n",
334 sbi->ll_md_exp->exp_obd->obd_name, buf);
335 OBD_FREE(buf, PAGE_SIZE);
336 GOTO(out_md_fid, err = -EPROTO);
339 size = sizeof(*data);
340 err = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_CONN_DATA),
341 KEY_CONN_DATA, &size, data);
343 CERROR("%s: Get connect data failed: rc = %d\n",
344 sbi->ll_md_exp->exp_obd->obd_name, err);
345 GOTO(out_md_fid, err);
348 LASSERT(osfs->os_bsize);
349 sb->s_blocksize = osfs->os_bsize;
350 sb->s_blocksize_bits = log2(osfs->os_bsize);
351 sb->s_magic = LL_SUPER_MAGIC;
352 sb->s_maxbytes = MAX_LFS_FILESIZE;
353 sbi->ll_namelen = osfs->os_namelen;
354 sbi->ll_mnt.mnt = current->fs->root.mnt;
356 if ((sbi->ll_flags & LL_SBI_USER_XATTR) &&
357 !(data->ocd_connect_flags & OBD_CONNECT_XATTR)) {
358 LCONSOLE_INFO("Disabling user_xattr feature because "
359 "it is not supported on the server\n");
360 sbi->ll_flags &= ~LL_SBI_USER_XATTR;
363 if (data->ocd_connect_flags & OBD_CONNECT_ACL) {
365 sb->s_flags |= MS_POSIXACL;
367 sbi->ll_flags |= LL_SBI_ACL;
369 LCONSOLE_INFO("client wants to enable acl, but mdt not!\n");
371 sb->s_flags &= ~MS_POSIXACL;
373 sbi->ll_flags &= ~LL_SBI_ACL;
376 if (data->ocd_connect_flags & OBD_CONNECT_64BITHASH)
377 sbi->ll_flags |= LL_SBI_64BIT_HASH;
379 if (data->ocd_connect_flags & OBD_CONNECT_LAYOUTLOCK)
380 sbi->ll_flags |= LL_SBI_LAYOUT_LOCK;
382 if (obd_connect_has_secctx(data))
383 sbi->ll_flags |= LL_SBI_FILE_SECCTX;
385 if (data->ocd_ibits_known & MDS_INODELOCK_XATTR) {
386 if (!(data->ocd_connect_flags & OBD_CONNECT_MAX_EASIZE)) {
387 LCONSOLE_INFO("%s: disabling xattr cache due to "
388 "unknown maximum xattr size.\n", dt);
389 } else if (!sbi->ll_xattr_cache_set) {
390 /* If xattr_cache is already set (no matter 0 or 1)
391 * during processing llog, it won't be enabled here. */
392 sbi->ll_flags |= LL_SBI_XATTR_CACHE;
393 sbi->ll_xattr_cache_enabled = 1;
397 sbi->ll_dt_obd = class_name2obd(dt);
398 if (!sbi->ll_dt_obd) {
399 CERROR("DT %s: not setup or attached\n", dt);
400 GOTO(out_md_fid, err = -ENODEV);
403 /* pass client page size via ocd_grant_blkbits, the server should report
404 * back its backend blocksize for grant calculation purpose */
405 data->ocd_grant_blkbits = PAGE_SHIFT;
407 /* indicate OST features supported by this client */
408 data->ocd_connect_flags = OBD_CONNECT_GRANT | OBD_CONNECT_VERSION |
409 OBD_CONNECT_REQPORTAL | OBD_CONNECT_BRW_SIZE |
410 OBD_CONNECT_CANCELSET | OBD_CONNECT_FID |
411 OBD_CONNECT_SRVLOCK | OBD_CONNECT_TRUNCLOCK|
412 OBD_CONNECT_AT | OBD_CONNECT_OSS_CAPA |
413 OBD_CONNECT_VBR | OBD_CONNECT_FULL20 |
414 OBD_CONNECT_64BITHASH | OBD_CONNECT_MAXBYTES |
415 OBD_CONNECT_EINPROGRESS |
416 OBD_CONNECT_JOBSTATS | OBD_CONNECT_LVB_TYPE |
417 OBD_CONNECT_LAYOUTLOCK |
418 OBD_CONNECT_PINGLESS | OBD_CONNECT_LFSCK |
419 OBD_CONNECT_BULK_MBITS | OBD_CONNECT_SHORTIO |
420 OBD_CONNECT_FLAGS2 | OBD_CONNECT_GRANT_SHRINK;
422 /* The client currently advertises support for OBD_CONNECT_LOCKAHEAD_OLD so it
423 * can interoperate with an older version of lockahead which was released prior
424 * to landing in master. This support will be dropped when 2.13 development
425 * starts. At the point, we should not just drop the connect flag (below), we
426 * should also remove the support in the code.
428 * Removing it means a few things:
429 * 1. Remove this section here
430 * 2. Remove CEF_NONBLOCK in ll_file_lockahead()
431 * 3. Remove function exp_connect_lockahead_old
432 * 4. Remove LDLM_FL_LOCKAHEAD_OLD_RESERVED in lustre_dlm_flags.h
434 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 12, 50, 0)
435 data->ocd_connect_flags |= OBD_CONNECT_LOCKAHEAD_OLD;
438 data->ocd_connect_flags2 = OBD_CONNECT2_LOCKAHEAD;
440 if (!OBD_FAIL_CHECK(OBD_FAIL_OSC_CONNECT_GRANT_PARAM))
441 data->ocd_connect_flags |= OBD_CONNECT_GRANT_PARAM;
443 /* OBD_CONNECT_CKSUM should always be set, even if checksums are
444 * disabled by default, because it can still be enabled on the
445 * fly via /sys. As a consequence, we still need to come to an
446 * agreement on the supported algorithms at connect time
448 data->ocd_connect_flags |= OBD_CONNECT_CKSUM;
450 if (OBD_FAIL_CHECK(OBD_FAIL_OSC_CKSUM_ADLER_ONLY))
451 data->ocd_cksum_types = OBD_CKSUM_ADLER;
453 data->ocd_cksum_types = obd_cksum_types_supported_client();
455 #ifdef HAVE_LRU_RESIZE_SUPPORT
456 data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
458 /* always ping even if server suppress_pings */
459 if (sbi->ll_flags & LL_SBI_ALWAYS_PING)
460 data->ocd_connect_flags &= ~OBD_CONNECT_PINGLESS;
462 CDEBUG(D_RPCTRACE, "ocd_connect_flags: %#llx ocd_version: %d "
463 "ocd_grant: %d\n", data->ocd_connect_flags,
464 data->ocd_version, data->ocd_grant);
466 sbi->ll_dt_obd->obd_upcall.onu_owner = &sbi->ll_lco;
467 sbi->ll_dt_obd->obd_upcall.onu_upcall = cl_ocd_update;
469 data->ocd_brw_size = DT_MAX_BRW_SIZE;
471 err = obd_connect(NULL, &sbi->ll_dt_exp, sbi->ll_dt_obd,
472 &sbi->ll_sb_uuid, data, sbi->ll_cache);
474 LCONSOLE_ERROR_MSG(0x150, "An OST (dt %s) is performing "
475 "recovery, of which this client is not a "
476 "part. Please wait for recovery to "
477 "complete, abort, or time out.\n", dt);
480 CERROR("%s: Cannot connect to %s: rc = %d\n",
481 sbi->ll_dt_exp->exp_obd->obd_name, dt, err);
485 sbi->ll_dt_exp->exp_connect_data = *data;
487 /* Don't change value if it was specified in the config log */
488 if (sbi->ll_ra_info.ra_max_read_ahead_whole_pages == -1)
489 sbi->ll_ra_info.ra_max_read_ahead_whole_pages =
490 max_t(unsigned long, SBI_DEFAULT_READAHEAD_WHOLE_MAX,
491 (data->ocd_brw_size >> PAGE_SHIFT));
493 err = obd_fid_init(sbi->ll_dt_exp->exp_obd, sbi->ll_dt_exp,
494 LUSTRE_SEQ_METADATA);
496 CERROR("%s: Can't init data layer FID infrastructure, "
497 "rc = %d\n", sbi->ll_dt_exp->exp_obd->obd_name, err);
501 mutex_lock(&sbi->ll_lco.lco_lock);
502 sbi->ll_lco.lco_flags = data->ocd_connect_flags;
503 sbi->ll_lco.lco_md_exp = sbi->ll_md_exp;
504 sbi->ll_lco.lco_dt_exp = sbi->ll_dt_exp;
505 mutex_unlock(&sbi->ll_lco.lco_lock);
507 fid_zero(&sbi->ll_root_fid);
508 err = md_get_root(sbi->ll_md_exp, get_mount_fileset(sb),
511 CERROR("cannot mds_connect: rc = %d\n", err);
512 GOTO(out_lock_cn_cb, err);
514 if (!fid_is_sane(&sbi->ll_root_fid)) {
515 CERROR("%s: Invalid root fid "DFID" during mount\n",
516 sbi->ll_md_exp->exp_obd->obd_name,
517 PFID(&sbi->ll_root_fid));
518 GOTO(out_lock_cn_cb, err = -EINVAL);
520 CDEBUG(D_SUPER, "rootfid "DFID"\n", PFID(&sbi->ll_root_fid));
522 sb->s_op = &lustre_super_operations;
523 #ifdef HAVE_XATTR_HANDLER_FLAGS
524 sb->s_xattr = ll_xattr_handlers;
526 #if THREAD_SIZE >= 8192 /*b=17630*/
527 sb->s_export_op = &lustre_export_operations;
531 * XXX: move this to after cbd setup? */
532 valid = OBD_MD_FLGETATTR | OBD_MD_FLBLOCKS | OBD_MD_FLMODEASIZE;
533 if (sbi->ll_flags & LL_SBI_ACL)
534 valid |= OBD_MD_FLACL;
536 OBD_ALLOC_PTR(op_data);
538 GOTO(out_lock_cn_cb, err = -ENOMEM);
540 op_data->op_fid1 = sbi->ll_root_fid;
541 op_data->op_mode = 0;
542 op_data->op_valid = valid;
544 err = md_getattr(sbi->ll_md_exp, op_data, &request);
546 OBD_FREE_PTR(op_data);
548 CERROR("%s: md_getattr failed for root: rc = %d\n",
549 sbi->ll_md_exp->exp_obd->obd_name, err);
550 GOTO(out_lock_cn_cb, err);
553 err = md_get_lustre_md(sbi->ll_md_exp, request, sbi->ll_dt_exp,
554 sbi->ll_md_exp, &lmd);
556 CERROR("failed to understand root inode md: rc = %d\n", err);
557 ptlrpc_req_finished(request);
558 GOTO(out_lock_cn_cb, err);
561 LASSERT(fid_is_sane(&sbi->ll_root_fid));
562 root = ll_iget(sb, cl_fid_build_ino(&sbi->ll_root_fid,
563 sbi->ll_flags & LL_SBI_32BIT_API),
565 md_free_lustre_md(sbi->ll_md_exp, &lmd);
566 ptlrpc_req_finished(request);
569 #ifdef CONFIG_FS_POSIX_ACL
571 posix_acl_release(lmd.posix_acl);
572 lmd.posix_acl = NULL;
575 err = IS_ERR(root) ? PTR_ERR(root) : -EBADF;
577 CERROR("lustre_lite: bad iget4 for root\n");
581 checksum = sbi->ll_flags & LL_SBI_CHECKSUM;
582 if (sbi->ll_checksum_set) {
583 err = obd_set_info_async(NULL, sbi->ll_dt_exp,
584 sizeof(KEY_CHECKSUM), KEY_CHECKSUM,
585 sizeof(checksum), &checksum, NULL);
587 CERROR("%s: Set checksum failed: rc = %d\n",
588 sbi->ll_dt_exp->exp_obd->obd_name, err);
594 sb->s_root = d_make_root(root);
595 if (sb->s_root == NULL) {
596 CERROR("%s: can't make root dentry\n",
597 ll_get_fsname(sb, NULL, 0));
598 GOTO(out_root, err = -ENOMEM);
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, ll_get_fsname(sb, NULL, 0), 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, ll_get_fsname(sb, NULL, 0), 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);
680 rc = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_MAX_EASIZE),
681 KEY_MAX_EASIZE, &size, lmmsize);
683 CERROR("Get max mdsize error rc %d\n", rc);
689 * Get the value of the default_easize parameter.
691 * \see client_obd::cl_default_mds_easize
693 * \param[in] sbi superblock info for this filesystem
694 * \param[out] lmmsize pointer to storage location for value
696 * \retval 0 on success
697 * \retval negative negated errno on failure
699 int ll_get_default_mdsize(struct ll_sb_info *sbi, int *lmmsize)
704 rc = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_DEFAULT_EASIZE),
705 KEY_DEFAULT_EASIZE, &size, lmmsize);
707 CERROR("Get default mdsize error rc %d\n", rc);
713 * Set the default_easize parameter to the given value.
715 * \see client_obd::cl_default_mds_easize
717 * \param[in] sbi superblock info for this filesystem
718 * \param[in] lmmsize the size to set
720 * \retval 0 on success
721 * \retval negative negated errno on failure
723 int ll_set_default_mdsize(struct ll_sb_info *sbi, int lmmsize)
727 if (lmmsize < sizeof(struct lov_mds_md) ||
728 lmmsize > OBD_MAX_DEFAULT_EA_SIZE)
731 rc = obd_set_info_async(NULL, sbi->ll_md_exp,
732 sizeof(KEY_DEFAULT_EASIZE), KEY_DEFAULT_EASIZE,
733 sizeof(int), &lmmsize, NULL);
738 static void client_common_put_super(struct super_block *sb)
740 struct ll_sb_info *sbi = ll_s2sbi(sb);
745 obd_fid_fini(sbi->ll_dt_exp->exp_obd);
746 obd_disconnect(sbi->ll_dt_exp);
747 sbi->ll_dt_exp = NULL;
749 ll_debugfs_unregister_super(sb);
751 obd_fid_fini(sbi->ll_md_exp->exp_obd);
752 obd_disconnect(sbi->ll_md_exp);
753 sbi->ll_md_exp = NULL;
758 void ll_kill_super(struct super_block *sb)
760 struct ll_sb_info *sbi;
764 if (!(sb->s_flags & MS_ACTIVE))
768 /* we need restore s_dev from changed for clustred NFS before put_super
769 * because new kernels have cached s_dev and change sb->s_dev in
770 * put_super not affected real removing devices */
772 sb->s_dev = sbi->ll_sdev_orig;
773 sbi->ll_umounting = 1;
775 /* wait running statahead threads to quit */
776 while (atomic_read(&sbi->ll_sa_running) > 0) {
777 set_current_state(TASK_UNINTERRUPTIBLE);
778 schedule_timeout(msecs_to_jiffies(MSEC_PER_SEC >> 3));
785 static inline int ll_set_opt(const char *opt, char *data, int fl)
787 if (strncmp(opt, data, strlen(opt)) != 0)
793 /* non-client-specific mount options are parsed in lmd_parse */
794 static int ll_options(char *options, struct ll_sb_info *sbi)
797 char *s1 = options, *s2;
798 int *flags = &sbi->ll_flags;
804 CDEBUG(D_CONFIG, "Parsing opts %s\n", options);
807 CDEBUG(D_SUPER, "next opt=%s\n", s1);
808 tmp = ll_set_opt("nolock", s1, LL_SBI_NOLCK);
813 tmp = ll_set_opt("flock", s1, LL_SBI_FLOCK);
818 tmp = ll_set_opt("localflock", s1, LL_SBI_LOCALFLOCK);
823 tmp = ll_set_opt("noflock", s1, LL_SBI_FLOCK|LL_SBI_LOCALFLOCK);
828 tmp = ll_set_opt("user_xattr", s1, LL_SBI_USER_XATTR);
833 tmp = ll_set_opt("nouser_xattr", s1, LL_SBI_USER_XATTR);
838 tmp = ll_set_opt("context", s1, 1);
841 tmp = ll_set_opt("fscontext", s1, 1);
844 tmp = ll_set_opt("defcontext", s1, 1);
847 tmp = ll_set_opt("rootcontext", s1, 1);
850 tmp = ll_set_opt("user_fid2path", s1, LL_SBI_USER_FID2PATH);
855 tmp = ll_set_opt("nouser_fid2path", s1, LL_SBI_USER_FID2PATH);
861 tmp = ll_set_opt("checksum", s1, LL_SBI_CHECKSUM);
864 sbi->ll_checksum_set = 1;
867 tmp = ll_set_opt("nochecksum", s1, LL_SBI_CHECKSUM);
870 sbi->ll_checksum_set = 1;
873 tmp = ll_set_opt("lruresize", s1, LL_SBI_LRU_RESIZE);
878 tmp = ll_set_opt("nolruresize", s1, LL_SBI_LRU_RESIZE);
883 tmp = ll_set_opt("lazystatfs", s1, LL_SBI_LAZYSTATFS);
888 tmp = ll_set_opt("nolazystatfs", s1, LL_SBI_LAZYSTATFS);
893 tmp = ll_set_opt("32bitapi", s1, LL_SBI_32BIT_API);
898 tmp = ll_set_opt("verbose", s1, LL_SBI_VERBOSE);
903 tmp = ll_set_opt("noverbose", s1, LL_SBI_VERBOSE);
908 tmp = ll_set_opt("always_ping", s1, LL_SBI_ALWAYS_PING);
913 LCONSOLE_ERROR_MSG(0x152, "Unknown option '%s', won't mount.\n",
919 s2 = strchr(s1, ',');
927 void ll_lli_init(struct ll_inode_info *lli)
929 lli->lli_inode_magic = LLI_INODE_MAGIC;
931 spin_lock_init(&lli->lli_lock);
932 lli->lli_posix_acl = NULL;
933 /* Do not set lli_fid, it has been initialized already. */
934 fid_zero(&lli->lli_pfid);
935 lli->lli_mds_read_och = NULL;
936 lli->lli_mds_write_och = NULL;
937 lli->lli_mds_exec_och = NULL;
938 lli->lli_open_fd_read_count = 0;
939 lli->lli_open_fd_write_count = 0;
940 lli->lli_open_fd_exec_count = 0;
941 mutex_init(&lli->lli_och_mutex);
942 spin_lock_init(&lli->lli_agl_lock);
943 spin_lock_init(&lli->lli_layout_lock);
944 ll_layout_version_set(lli, CL_LAYOUT_GEN_NONE);
945 lli->lli_clob = NULL;
947 init_rwsem(&lli->lli_xattrs_list_rwsem);
948 mutex_init(&lli->lli_xattrs_enq_lock);
950 LASSERT(lli->lli_vfs_inode.i_mode != 0);
951 if (S_ISDIR(lli->lli_vfs_inode.i_mode)) {
952 mutex_init(&lli->lli_readdir_mutex);
953 lli->lli_opendir_key = NULL;
955 spin_lock_init(&lli->lli_sa_lock);
956 lli->lli_opendir_pid = 0;
957 lli->lli_sa_enabled = 0;
958 lli->lli_def_stripe_offset = -1;
959 init_rwsem(&lli->lli_lsm_sem);
961 mutex_init(&lli->lli_size_mutex);
962 lli->lli_symlink_name = NULL;
963 init_rwsem(&lli->lli_trunc_sem);
964 range_lock_tree_init(&lli->lli_write_tree);
965 init_rwsem(&lli->lli_glimpse_sem);
966 lli->lli_glimpse_time = ktime_set(0, 0);
967 INIT_LIST_HEAD(&lli->lli_agl_list);
968 lli->lli_agl_index = 0;
969 lli->lli_async_rc = 0;
970 spin_lock_init(&lli->lli_heat_lock);
971 obd_heat_clear(lli->lli_heat_instances, OBD_HEAT_COUNT);
972 lli->lli_heat_flags = 0;
974 mutex_init(&lli->lli_layout_mutex);
975 memset(lli->lli_jobid, 0, sizeof(lli->lli_jobid));
978 #define MAX_STRING_SIZE 128
980 #ifndef HAVE_SUPER_SETUP_BDI_NAME
982 #define LSI_BDI_INITIALIZED 0x00400000
984 #ifndef HAVE_BDI_CAP_MAP_COPY
985 # define BDI_CAP_MAP_COPY 0
988 static int super_setup_bdi_name(struct super_block *sb, char *fmt, ...)
990 struct lustre_sb_info *lsi = s2lsi(sb);
991 char buf[MAX_STRING_SIZE];
995 err = bdi_init(&lsi->lsi_bdi);
999 lsi->lsi_flags |= LSI_BDI_INITIALIZED;
1000 lsi->lsi_bdi.capabilities = BDI_CAP_MAP_COPY;
1001 lsi->lsi_bdi.name = "lustre";
1002 va_start(args, fmt);
1003 vsnprintf(buf, MAX_STRING_SIZE, fmt, args);
1005 err = bdi_register(&lsi->lsi_bdi, NULL, "%s", buf);
1008 sb->s_bdi = &lsi->lsi_bdi;
1012 #endif /* !HAVE_SUPER_SETUP_BDI_NAME */
1014 int ll_fill_super(struct super_block *sb, struct vfsmount *mnt)
1016 struct lustre_profile *lprof = NULL;
1017 struct lustre_sb_info *lsi = s2lsi(sb);
1018 struct ll_sb_info *sbi = NULL;
1019 char *dt = NULL, *md = NULL;
1020 char *profilenm = get_profile_name(sb);
1021 struct config_llog_instance *cfg;
1022 /* %p for void* in printf needs 16+2 characters: 0xffffffffffffffff */
1023 const int instlen = 16 + 2;
1024 unsigned long cfg_instance = ll_get_cfg_instance(sb);
1025 char name[MAX_STRING_SIZE];
1033 /* for ASLR, to map between cfg_instance and hashed ptr */
1034 CDEBUG(D_VFSTRACE, "VFS Op: cfg_instance %s-%016lx (sb %p)\n",
1035 profilenm, cfg_instance, sb);
1037 try_module_get(THIS_MODULE);
1041 GOTO(out_free_cfg, err = -ENOMEM);
1043 /* client additional sb info */
1044 lsi->lsi_llsbi = sbi = ll_init_sbi();
1046 GOTO(out_free_cfg, err = -ENOMEM);
1048 err = ll_options(lsi->lsi_lmd->lmd_opts, sbi);
1050 GOTO(out_free_cfg, err);
1052 err = super_setup_bdi_name(sb, "lustre-%016lx", cfg_instance);
1054 GOTO(out_free_cfg, err);
1056 #ifndef HAVE_DCACHE_LOCK
1057 /* kernel >= 2.6.38 store dentry operations in sb->s_d_op. */
1058 sb->s_d_op = &ll_d_ops;
1061 len = strlen(profilenm);
1062 ptr = strrchr(profilenm, '-');
1063 if (ptr && (strcmp(ptr, "-client") == 0))
1067 snprintf(name, MAX_STRING_SIZE, "%.*s-%016lx", len,
1068 profilenm, cfg_instance);
1070 /* Call ll_debugfs_register_super() before lustre_process_log()
1071 * so that "llite.*.*" params can be processed correctly.
1073 err = ll_debugfs_register_super(sb, name);
1075 CERROR("%s: could not register mountpoint in llite: rc = %d\n",
1076 ll_get_fsname(sb, NULL, 0), err);
1080 /* The cfg_instance is a value unique to this super, in case some
1081 * joker tries to mount the same fs at two mount points.
1083 cfg->cfg_instance = cfg_instance;
1084 cfg->cfg_uuid = lsi->lsi_llsbi->ll_sb_uuid;
1085 cfg->cfg_callback = class_config_llog_handler;
1086 cfg->cfg_sub_clds = CONFIG_SUB_CLIENT;
1087 /* set up client obds */
1088 err = lustre_process_log(sb, profilenm, cfg);
1090 GOTO(out_debugfs, err);
1092 /* Profile set with LCFG_MOUNTOPT so we can find our mdc and osc obds */
1093 lprof = class_get_profile(profilenm);
1094 if (lprof == NULL) {
1095 LCONSOLE_ERROR_MSG(0x156, "The client profile '%s' could not be"
1096 " read from the MGS. Does that filesystem "
1097 "exist?\n", profilenm);
1098 GOTO(out_debugfs, err = -EINVAL);
1100 CDEBUG(D_CONFIG, "Found profile %s: mdc=%s osc=%s\n", profilenm,
1101 lprof->lp_md, lprof->lp_dt);
1103 dt_len = strlen(lprof->lp_dt) + instlen + 2;
1104 OBD_ALLOC(dt, dt_len);
1106 GOTO(out_profile, err = -ENOMEM);
1107 snprintf(dt, dt_len - 1, "%s-%016lx", lprof->lp_dt, cfg_instance);
1109 md_len = strlen(lprof->lp_md) + instlen + 2;
1110 OBD_ALLOC(md, md_len);
1112 GOTO(out_free_dt, err = -ENOMEM);
1113 snprintf(md, md_len - 1, "%s-%016lx", lprof->lp_md, cfg_instance);
1115 /* connections, registrations, sb setup */
1116 err = client_common_fill_super(sb, md, dt, mnt);
1118 GOTO(out_free_md, err);
1120 sbi->ll_client_common_fill_super_succeeded = 1;
1124 OBD_FREE(md, md_len);
1127 OBD_FREE(dt, dt_len);
1130 class_put_profile(lprof);
1133 ll_debugfs_unregister_super(sb);
1140 else if (sbi->ll_flags & LL_SBI_VERBOSE)
1141 LCONSOLE_WARN("Mounted %s\n", profilenm);
1143 } /* ll_fill_super */
1145 void ll_put_super(struct super_block *sb)
1147 struct config_llog_instance cfg, params_cfg;
1148 struct obd_device *obd;
1149 struct lustre_sb_info *lsi = s2lsi(sb);
1150 struct ll_sb_info *sbi = ll_s2sbi(sb);
1151 char *profilenm = get_profile_name(sb);
1152 unsigned long cfg_instance = ll_get_cfg_instance(sb);
1154 int next, force = 1, rc = 0;
1158 GOTO(out_no_sbi, 0);
1160 /* Should replace instance_id with something better for ASLR */
1161 CDEBUG(D_VFSTRACE, "VFS Op: cfg_instance %s-%016lx (sb %p)\n",
1162 profilenm, cfg_instance, sb);
1164 cfg.cfg_instance = cfg_instance;
1165 lustre_end_log(sb, profilenm, &cfg);
1167 params_cfg.cfg_instance = cfg_instance;
1168 lustre_end_log(sb, PARAMS_FILENAME, ¶ms_cfg);
1170 if (sbi->ll_md_exp) {
1171 obd = class_exp2obd(sbi->ll_md_exp);
1173 force = obd->obd_force;
1176 /* Wait for unstable pages to be committed to stable storage */
1178 struct l_wait_info lwi = LWI_INTR(LWI_ON_SIGNAL_NOOP, NULL);
1179 rc = l_wait_event(sbi->ll_cache->ccc_unstable_waitq,
1180 atomic_long_read(&sbi->ll_cache->ccc_unstable_nr) == 0,
1184 ccc_count = atomic_long_read(&sbi->ll_cache->ccc_unstable_nr);
1185 if (force == 0 && rc != -EINTR)
1186 LASSERTF(ccc_count == 0, "count: %li\n", ccc_count);
1188 /* We need to set force before the lov_disconnect in
1189 lustre_common_put_super, since l_d cleans up osc's as well. */
1192 while ((obd = class_devices_in_group(&sbi->ll_sb_uuid,
1194 obd->obd_force = force;
1198 if (sbi->ll_client_common_fill_super_succeeded) {
1199 /* Only if client_common_fill_super succeeded */
1200 client_common_put_super(sb);
1204 while ((obd = class_devices_in_group(&sbi->ll_sb_uuid, &next)) !=NULL) {
1205 class_manual_cleanup(obd);
1208 if (sbi->ll_flags & LL_SBI_VERBOSE)
1209 LCONSOLE_WARN("Unmounted %s\n", profilenm ? profilenm : "");
1212 class_del_profile(profilenm);
1214 #ifndef HAVE_SUPER_SETUP_BDI_NAME
1215 if (lsi->lsi_flags & LSI_BDI_INITIALIZED) {
1216 bdi_destroy(&lsi->lsi_bdi);
1217 lsi->lsi_flags &= ~LSI_BDI_INITIALIZED;
1222 lsi->lsi_llsbi = NULL;
1224 lustre_common_put_super(sb);
1226 cl_env_cache_purge(~0);
1228 module_put(THIS_MODULE);
1231 } /* client_put_super */
1233 struct inode *ll_inode_from_resource_lock(struct ldlm_lock *lock)
1235 struct inode *inode = NULL;
1237 /* NOTE: we depend on atomic igrab() -bzzz */
1238 lock_res_and_lock(lock);
1239 if (lock->l_resource->lr_lvb_inode) {
1240 struct ll_inode_info * lli;
1241 lli = ll_i2info(lock->l_resource->lr_lvb_inode);
1242 if (lli->lli_inode_magic == LLI_INODE_MAGIC) {
1243 inode = igrab(lock->l_resource->lr_lvb_inode);
1245 inode = lock->l_resource->lr_lvb_inode;
1246 LDLM_DEBUG_LIMIT(inode->i_state & I_FREEING ? D_INFO :
1247 D_WARNING, lock, "lr_lvb_inode %p is "
1248 "bogus: magic %08x",
1249 lock->l_resource->lr_lvb_inode,
1250 lli->lli_inode_magic);
1254 unlock_res_and_lock(lock);
1258 void ll_dir_clear_lsm_md(struct inode *inode)
1260 struct ll_inode_info *lli = ll_i2info(inode);
1262 LASSERT(S_ISDIR(inode->i_mode));
1264 if (lli->lli_lsm_md != NULL) {
1265 lmv_free_memmd(lli->lli_lsm_md);
1266 lli->lli_lsm_md = NULL;
1270 static struct inode *ll_iget_anon_dir(struct super_block *sb,
1271 const struct lu_fid *fid,
1272 struct lustre_md *md)
1274 struct ll_sb_info *sbi = ll_s2sbi(sb);
1275 struct mdt_body *body = md->body;
1276 struct inode *inode;
1280 ino = cl_fid_build_ino(fid, sbi->ll_flags & LL_SBI_32BIT_API);
1281 inode = iget_locked(sb, ino);
1282 if (inode == NULL) {
1283 CERROR("%s: failed get simple inode "DFID": rc = -ENOENT\n",
1284 ll_get_fsname(sb, NULL, 0), PFID(fid));
1285 RETURN(ERR_PTR(-ENOENT));
1288 if (inode->i_state & I_NEW) {
1289 struct ll_inode_info *lli = ll_i2info(inode);
1290 struct lmv_stripe_md *lsm = md->lmv;
1292 inode->i_mode = (inode->i_mode & ~S_IFMT) |
1293 (body->mbo_mode & S_IFMT);
1294 LASSERTF(S_ISDIR(inode->i_mode), "Not slave inode "DFID"\n",
1297 inode->i_mtime.tv_sec = 0;
1298 inode->i_atime.tv_sec = 0;
1299 inode->i_ctime.tv_sec = 0;
1302 #ifdef HAVE_BACKING_DEV_INFO
1303 /* initializing backing dev info. */
1304 inode->i_mapping->backing_dev_info =
1305 &s2lsi(inode->i_sb)->lsi_bdi;
1307 inode->i_op = &ll_dir_inode_operations;
1308 inode->i_fop = &ll_dir_operations;
1309 lli->lli_fid = *fid;
1312 LASSERT(lsm != NULL);
1313 /* master object FID */
1314 lli->lli_pfid = body->mbo_fid1;
1315 CDEBUG(D_INODE, "lli %p slave "DFID" master "DFID"\n",
1316 lli, PFID(fid), PFID(&lli->lli_pfid));
1317 unlock_new_inode(inode);
1323 static int ll_init_lsm_md(struct inode *inode, struct lustre_md *md)
1326 struct lmv_stripe_md *lsm = md->lmv;
1327 struct ll_inode_info *lli = ll_i2info(inode);
1330 LASSERT(lsm != NULL);
1332 CDEBUG(D_INODE, "%s: "DFID" set dir layout:\n",
1333 ll_get_fsname(inode->i_sb, NULL, 0),
1334 PFID(&lli->lli_fid));
1335 lsm_md_dump(D_INODE, lsm);
1337 /* XXX sigh, this lsm_root initialization should be in
1338 * LMV layer, but it needs ll_iget right now, so we
1339 * put this here right now. */
1340 for (i = 0; i < lsm->lsm_md_stripe_count; i++) {
1341 fid = &lsm->lsm_md_oinfo[i].lmo_fid;
1342 LASSERT(lsm->lsm_md_oinfo[i].lmo_root == NULL);
1343 /* Unfortunately ll_iget will call ll_update_inode,
1344 * where the initialization of slave inode is slightly
1345 * different, so it reset lsm_md to NULL to avoid
1346 * initializing lsm for slave inode. */
1347 lsm->lsm_md_oinfo[i].lmo_root =
1348 ll_iget_anon_dir(inode->i_sb, fid, md);
1349 if (IS_ERR(lsm->lsm_md_oinfo[i].lmo_root)) {
1350 int rc = PTR_ERR(lsm->lsm_md_oinfo[i].lmo_root);
1352 lsm->lsm_md_oinfo[i].lmo_root = NULL;
1354 iput(lsm->lsm_md_oinfo[i].lmo_root);
1355 lsm->lsm_md_oinfo[i].lmo_root = NULL;
1361 lli->lli_lsm_md = lsm;
1366 static int ll_update_lsm_md(struct inode *inode, struct lustre_md *md)
1368 struct ll_inode_info *lli = ll_i2info(inode);
1369 struct lmv_stripe_md *lsm = md->lmv;
1374 LASSERT(S_ISDIR(inode->i_mode));
1375 CDEBUG(D_INODE, "update lsm %p of "DFID"\n", lli->lli_lsm_md,
1376 PFID(ll_inode2fid(inode)));
1379 * no striped information from request, lustre_md from req does not
1380 * include stripeEA, see ll_md_setattr()
1386 * normally dir layout doesn't change, only take read lock to check
1387 * that to avoid blocking other MD operations.
1389 if (lli->lli_lsm_md)
1390 down_read(&lli->lli_lsm_sem);
1392 down_write(&lli->lli_lsm_sem);
1395 * if dir layout mismatch, check whether version is increased, which
1396 * means layout is changed, this happens in dir migration and lfsck.
1398 if (lli->lli_lsm_md && !lsm_md_eq(lli->lli_lsm_md, lsm)) {
1399 if (lsm->lsm_md_layout_version <=
1400 lli->lli_lsm_md->lsm_md_layout_version) {
1401 CERROR("%s: "DFID" dir layout mismatch:\n",
1402 ll_get_fsname(inode->i_sb, NULL, 0),
1403 PFID(&lli->lli_fid));
1404 lsm_md_dump(D_ERROR, lli->lli_lsm_md);
1405 lsm_md_dump(D_ERROR, lsm);
1406 GOTO(unlock, rc = -EINVAL);
1409 /* layout changed, switch to write lock */
1410 up_read(&lli->lli_lsm_sem);
1411 down_write(&lli->lli_lsm_sem);
1412 ll_dir_clear_lsm_md(inode);
1415 /* set directory layout */
1416 if (!lli->lli_lsm_md) {
1417 struct cl_attr *attr;
1419 rc = ll_init_lsm_md(inode, md);
1420 up_write(&lli->lli_lsm_sem);
1424 /* set md->lmv to NULL, so the following free lustre_md
1425 * will not free this lsm */
1429 * md_merge_attr() may take long, since lsm is already set,
1430 * switch to read lock.
1432 down_read(&lli->lli_lsm_sem);
1434 OBD_ALLOC_PTR(attr);
1436 GOTO(unlock, rc = -ENOMEM);
1438 /* validate the lsm */
1439 rc = md_merge_attr(ll_i2mdexp(inode), lsm, attr,
1440 ll_md_blocking_ast);
1446 if (md->body->mbo_valid & OBD_MD_FLNLINK)
1447 md->body->mbo_nlink = attr->cat_nlink;
1448 if (md->body->mbo_valid & OBD_MD_FLSIZE)
1449 md->body->mbo_size = attr->cat_size;
1450 if (md->body->mbo_valid & OBD_MD_FLATIME)
1451 md->body->mbo_atime = attr->cat_atime;
1452 if (md->body->mbo_valid & OBD_MD_FLCTIME)
1453 md->body->mbo_ctime = attr->cat_ctime;
1454 if (md->body->mbo_valid & OBD_MD_FLMTIME)
1455 md->body->mbo_mtime = attr->cat_mtime;
1460 up_read(&lli->lli_lsm_sem);
1465 void ll_clear_inode(struct inode *inode)
1467 struct ll_inode_info *lli = ll_i2info(inode);
1468 struct ll_sb_info *sbi = ll_i2sbi(inode);
1471 CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p)\n",
1472 PFID(ll_inode2fid(inode)), inode);
1474 if (S_ISDIR(inode->i_mode)) {
1475 /* these should have been cleared in ll_file_release */
1476 LASSERT(lli->lli_opendir_key == NULL);
1477 LASSERT(lli->lli_sai == NULL);
1478 LASSERT(lli->lli_opendir_pid == 0);
1481 md_null_inode(sbi->ll_md_exp, ll_inode2fid(inode));
1483 LASSERT(!lli->lli_open_fd_write_count);
1484 LASSERT(!lli->lli_open_fd_read_count);
1485 LASSERT(!lli->lli_open_fd_exec_count);
1487 if (lli->lli_mds_write_och)
1488 ll_md_real_close(inode, FMODE_WRITE);
1489 if (lli->lli_mds_exec_och)
1490 ll_md_real_close(inode, FMODE_EXEC);
1491 if (lli->lli_mds_read_och)
1492 ll_md_real_close(inode, FMODE_READ);
1494 if (S_ISLNK(inode->i_mode) && lli->lli_symlink_name) {
1495 OBD_FREE(lli->lli_symlink_name,
1496 strlen(lli->lli_symlink_name) + 1);
1497 lli->lli_symlink_name = NULL;
1500 ll_xattr_cache_destroy(inode);
1502 #ifdef CONFIG_FS_POSIX_ACL
1503 forget_all_cached_acls(inode);
1504 if (lli->lli_posix_acl) {
1505 posix_acl_release(lli->lli_posix_acl);
1506 lli->lli_posix_acl = NULL;
1509 lli->lli_inode_magic = LLI_INODE_DEAD;
1511 if (S_ISDIR(inode->i_mode))
1512 ll_dir_clear_lsm_md(inode);
1513 else if (S_ISREG(inode->i_mode) && !is_bad_inode(inode))
1514 LASSERT(list_empty(&lli->lli_agl_list));
1517 * XXX This has to be done before lsm is freed below, because
1518 * cl_object still uses inode lsm.
1520 cl_inode_fini(inode);
1525 static int ll_md_setattr(struct dentry *dentry, struct md_op_data *op_data)
1527 struct lustre_md md;
1528 struct inode *inode = dentry->d_inode;
1529 struct ll_sb_info *sbi = ll_i2sbi(inode);
1530 struct ptlrpc_request *request = NULL;
1534 op_data = ll_prep_md_op_data(op_data, inode, NULL, NULL, 0, 0,
1535 LUSTRE_OPC_ANY, NULL);
1536 if (IS_ERR(op_data))
1537 RETURN(PTR_ERR(op_data));
1539 rc = md_setattr(sbi->ll_md_exp, op_data, NULL, 0, &request);
1541 ptlrpc_req_finished(request);
1542 if (rc == -ENOENT) {
1544 /* Unlinked special device node? Or just a race?
1545 * Pretend we done everything. */
1546 if (!S_ISREG(inode->i_mode) &&
1547 !S_ISDIR(inode->i_mode)) {
1548 ia_valid = op_data->op_attr.ia_valid;
1549 op_data->op_attr.ia_valid &= ~TIMES_SET_FLAGS;
1550 rc = simple_setattr(dentry, &op_data->op_attr);
1551 op_data->op_attr.ia_valid = ia_valid;
1553 } else if (rc != -EPERM && rc != -EACCES && rc != -ETXTBSY) {
1554 CERROR("md_setattr fails: rc = %d\n", rc);
1559 rc = md_get_lustre_md(sbi->ll_md_exp, request, sbi->ll_dt_exp,
1560 sbi->ll_md_exp, &md);
1562 ptlrpc_req_finished(request);
1566 ia_valid = op_data->op_attr.ia_valid;
1567 /* inode size will be in ll_setattr_ost, can't do it now since dirty
1568 * cache is not cleared yet. */
1569 op_data->op_attr.ia_valid &= ~(TIMES_SET_FLAGS | ATTR_SIZE);
1570 if (S_ISREG(inode->i_mode))
1572 rc = simple_setattr(dentry, &op_data->op_attr);
1573 if (S_ISREG(inode->i_mode))
1574 inode_unlock(inode);
1575 op_data->op_attr.ia_valid = ia_valid;
1577 rc = ll_update_inode(inode, &md);
1578 ptlrpc_req_finished(request);
1583 /* If this inode has objects allocated to it (lsm != NULL), then the OST
1584 * object(s) determine the file size and mtime. Otherwise, the MDS will
1585 * keep these values until such a time that objects are allocated for it.
1586 * We do the MDS operations first, as it is checking permissions for us.
1587 * We don't to the MDS RPC if there is nothing that we want to store there,
1588 * otherwise there is no harm in updating mtime/atime on the MDS if we are
1589 * going to do an RPC anyways.
1591 * If we are doing a truncate, we will send the mtime and ctime updates
1592 * to the OST with the punch RPC, otherwise we do an explicit setattr RPC.
1593 * I don't believe it is possible to get e.g. ATTR_MTIME_SET and ATTR_SIZE
1596 * In case of HSMimport, we only set attr on MDS.
1598 int ll_setattr_raw(struct dentry *dentry, struct iattr *attr,
1599 enum op_xvalid xvalid, bool hsm_import)
1601 struct inode *inode = dentry->d_inode;
1602 struct ll_inode_info *lli = ll_i2info(inode);
1603 struct md_op_data *op_data = NULL;
1607 CDEBUG(D_VFSTRACE, "%s: setattr inode "DFID"(%p) from %llu to %llu, "
1608 "valid %x, hsm_import %d\n",
1609 ll_get_fsname(inode->i_sb, NULL, 0), PFID(&lli->lli_fid),
1610 inode, i_size_read(inode), attr->ia_size, attr->ia_valid,
1613 if (attr->ia_valid & ATTR_SIZE) {
1614 /* Check new size against VFS/VM file size limit and rlimit */
1615 rc = inode_newsize_ok(inode, attr->ia_size);
1619 /* The maximum Lustre file size is variable, based on the
1620 * OST maximum object size and number of stripes. This
1621 * needs another check in addition to the VFS check above. */
1622 if (attr->ia_size > ll_file_maxbytes(inode)) {
1623 CDEBUG(D_INODE,"file "DFID" too large %llu > %llu\n",
1624 PFID(&lli->lli_fid), attr->ia_size,
1625 ll_file_maxbytes(inode));
1629 attr->ia_valid |= ATTR_MTIME | ATTR_CTIME;
1632 /* POSIX: check before ATTR_*TIME_SET set (from inode_change_ok) */
1633 if (attr->ia_valid & TIMES_SET_FLAGS) {
1634 if ((!uid_eq(current_fsuid(), inode->i_uid)) &&
1635 !cfs_capable(CFS_CAP_FOWNER))
1639 /* We mark all of the fields "set" so MDS/OST does not re-set them */
1640 if (!(xvalid & OP_XVALID_CTIME_SET) &&
1641 (attr->ia_valid & ATTR_CTIME)) {
1642 attr->ia_ctime = current_time(inode);
1643 xvalid |= OP_XVALID_CTIME_SET;
1645 if (!(attr->ia_valid & ATTR_ATIME_SET) &&
1646 (attr->ia_valid & ATTR_ATIME)) {
1647 attr->ia_atime = current_time(inode);
1648 attr->ia_valid |= ATTR_ATIME_SET;
1650 if (!(attr->ia_valid & ATTR_MTIME_SET) &&
1651 (attr->ia_valid & ATTR_MTIME)) {
1652 attr->ia_mtime = current_time(inode);
1653 attr->ia_valid |= ATTR_MTIME_SET;
1656 if (attr->ia_valid & (ATTR_MTIME | ATTR_CTIME))
1657 CDEBUG(D_INODE, "setting mtime %lld, ctime %lld, now = %lld\n",
1658 (s64)attr->ia_mtime.tv_sec, (s64)attr->ia_ctime.tv_sec,
1659 ktime_get_real_seconds());
1661 if (S_ISREG(inode->i_mode)) {
1662 if (attr->ia_valid & ATTR_SIZE)
1663 inode_dio_write_done(inode);
1664 inode_unlock(inode);
1667 /* We always do an MDS RPC, even if we're only changing the size;
1668 * only the MDS knows whether truncate() should fail with -ETXTBUSY */
1670 OBD_ALLOC_PTR(op_data);
1671 if (op_data == NULL)
1672 GOTO(out, rc = -ENOMEM);
1674 if (!hsm_import && attr->ia_valid & ATTR_SIZE) {
1675 /* If we are changing file size, file content is
1676 * modified, flag it.
1678 xvalid |= OP_XVALID_OWNEROVERRIDE;
1679 op_data->op_bias |= MDS_DATA_MODIFIED;
1680 ll_file_clear_flag(lli, LLIF_DATA_MODIFIED);
1683 if (attr->ia_valid & ATTR_FILE) {
1684 struct ll_file_data *fd = LUSTRE_FPRIVATE(attr->ia_file);
1686 if (fd->fd_lease_och)
1687 op_data->op_bias |= MDS_TRUNC_KEEP_LEASE;
1690 op_data->op_attr = *attr;
1691 op_data->op_xvalid = xvalid;
1693 rc = ll_md_setattr(dentry, op_data);
1697 if (!S_ISREG(inode->i_mode) || hsm_import)
1700 if (attr->ia_valid & (ATTR_SIZE | ATTR_ATIME | ATTR_ATIME_SET |
1701 ATTR_MTIME | ATTR_MTIME_SET | ATTR_CTIME) ||
1702 xvalid & OP_XVALID_CTIME_SET) {
1703 /* For truncate and utimes sending attributes to OSTs, setting
1704 * mtime/atime to the past will be performed under PW [0:EOF]
1705 * extent lock (new_size:EOF for truncate). It may seem
1706 * excessive to send mtime/atime updates to OSTs when not
1707 * setting times to past, but it is necessary due to possible
1708 * time de-synchronization between MDT inode and OST objects
1710 rc = cl_setattr_ost(lli->lli_clob, attr, xvalid, 0);
1713 /* If the file was restored, it needs to set dirty flag.
1715 * We've already sent MDS_DATA_MODIFIED flag in
1716 * ll_md_setattr() for truncate. However, the MDT refuses to
1717 * set the HS_DIRTY flag on released files, so we have to set
1718 * it again if the file has been restored. Please check how
1719 * LLIF_DATA_MODIFIED is set in vvp_io_setattr_fini().
1721 * Please notice that if the file is not released, the previous
1722 * MDS_DATA_MODIFIED has taken effect and usually
1723 * LLIF_DATA_MODIFIED is not set(see vvp_io_setattr_fini()).
1724 * This way we can save an RPC for common open + trunc
1726 if (ll_file_test_and_clear_flag(lli, LLIF_DATA_MODIFIED)) {
1727 struct hsm_state_set hss = {
1728 .hss_valid = HSS_SETMASK,
1729 .hss_setmask = HS_DIRTY,
1733 rc2 = ll_hsm_state_set(inode, &hss);
1734 /* truncate and write can happen at the same time, so that
1735 * the file can be set modified even though the file is not
1736 * restored from released state, and ll_hsm_state_set() is
1737 * not applicable for the file, and rc2 < 0 is normal in this
1740 CDEBUG(D_INFO, DFID "HSM set dirty failed: rc2 = %d\n",
1741 PFID(ll_inode2fid(inode)), rc2);
1746 if (op_data != NULL)
1747 ll_finish_md_op_data(op_data);
1749 if (S_ISREG(inode->i_mode)) {
1751 if ((attr->ia_valid & ATTR_SIZE) && !hsm_import)
1752 inode_dio_wait(inode);
1753 /* Once we've got the i_mutex, it's safe to set the S_NOSEC
1754 * flag. ll_update_inode (called from ll_md_setattr), clears
1755 * inode flags, so there is a gap where S_NOSEC is not set.
1756 * This can cause a writer to take the i_mutex unnecessarily,
1757 * but this is safe to do and should be rare. */
1758 inode_has_no_xattr(inode);
1761 ll_stats_ops_tally(ll_i2sbi(inode), (attr->ia_valid & ATTR_SIZE) ?
1762 LPROC_LL_TRUNC : LPROC_LL_SETATTR, 1);
1767 int ll_setattr(struct dentry *de, struct iattr *attr)
1769 int mode = de->d_inode->i_mode;
1770 enum op_xvalid xvalid = 0;
1772 if ((attr->ia_valid & (ATTR_CTIME|ATTR_SIZE|ATTR_MODE)) ==
1773 (ATTR_CTIME|ATTR_SIZE|ATTR_MODE))
1774 xvalid |= OP_XVALID_OWNEROVERRIDE;
1776 if (((attr->ia_valid & (ATTR_MODE|ATTR_FORCE|ATTR_SIZE)) ==
1777 (ATTR_SIZE|ATTR_MODE)) &&
1778 (((mode & S_ISUID) && !(attr->ia_mode & S_ISUID)) ||
1779 (((mode & (S_ISGID|S_IXGRP)) == (S_ISGID|S_IXGRP)) &&
1780 !(attr->ia_mode & S_ISGID))))
1781 attr->ia_valid |= ATTR_FORCE;
1783 if ((attr->ia_valid & ATTR_MODE) &&
1785 !(attr->ia_mode & S_ISUID) &&
1786 !(attr->ia_valid & ATTR_KILL_SUID))
1787 attr->ia_valid |= ATTR_KILL_SUID;
1789 if ((attr->ia_valid & ATTR_MODE) &&
1790 ((mode & (S_ISGID|S_IXGRP)) == (S_ISGID|S_IXGRP)) &&
1791 !(attr->ia_mode & S_ISGID) &&
1792 !(attr->ia_valid & ATTR_KILL_SGID))
1793 attr->ia_valid |= ATTR_KILL_SGID;
1795 return ll_setattr_raw(de, attr, xvalid, false);
1798 int ll_statfs_internal(struct ll_sb_info *sbi, struct obd_statfs *osfs,
1801 struct obd_statfs obd_osfs = { 0 };
1806 max_age = ktime_get_seconds() - OBD_STATFS_CACHE_SECONDS;
1808 rc = obd_statfs(NULL, sbi->ll_md_exp, osfs, max_age, flags);
1812 osfs->os_type = LL_SUPER_MAGIC;
1814 CDEBUG(D_SUPER, "MDC blocks %llu/%llu objects %llu/%llu\n",
1815 osfs->os_bavail, osfs->os_blocks, osfs->os_ffree, osfs->os_files);
1817 if (osfs->os_state & OS_STATE_SUM)
1820 if (sbi->ll_flags & LL_SBI_LAZYSTATFS)
1821 flags |= OBD_STATFS_NODELAY;
1823 rc = obd_statfs(NULL, sbi->ll_dt_exp, &obd_osfs, max_age, flags);
1824 if (rc) /* Possibly a filesystem with no OSTs. Report MDT totals. */
1827 CDEBUG(D_SUPER, "OSC blocks %llu/%llu objects %llu/%llu\n",
1828 obd_osfs.os_bavail, obd_osfs.os_blocks, obd_osfs.os_ffree,
1831 osfs->os_bsize = obd_osfs.os_bsize;
1832 osfs->os_blocks = obd_osfs.os_blocks;
1833 osfs->os_bfree = obd_osfs.os_bfree;
1834 osfs->os_bavail = obd_osfs.os_bavail;
1836 /* If we have _some_ OSTs, but don't have as many free objects on the
1837 * OSTs as inodes on the MDTs, reduce the reported number of inodes
1838 * to compensate, so that the "inodes in use" number is correct.
1839 * This should be kept in sync with lod_statfs() behaviour.
1841 if (obd_osfs.os_files && obd_osfs.os_ffree < osfs->os_ffree) {
1842 osfs->os_files = (osfs->os_files - osfs->os_ffree) +
1844 osfs->os_ffree = obd_osfs.os_ffree;
1850 int ll_statfs(struct dentry *de, struct kstatfs *sfs)
1852 struct super_block *sb = de->d_sb;
1853 struct obd_statfs osfs;
1854 __u64 fsid = huge_encode_dev(sb->s_dev);
1857 CDEBUG(D_VFSTRACE, "VFS Op: at %llu jiffies\n", get_jiffies_64());
1858 ll_stats_ops_tally(ll_s2sbi(sb), LPROC_LL_STAFS, 1);
1860 /* Some amount of caching on the client is allowed */
1861 rc = ll_statfs_internal(ll_s2sbi(sb), &osfs, OBD_STATFS_SUM);
1865 statfs_unpack(sfs, &osfs);
1867 /* We need to downshift for all 32-bit kernels, because we can't
1868 * tell if the kernel is being called via sys_statfs64() or not.
1869 * Stop before overflowing f_bsize - in which case it is better
1870 * to just risk EOVERFLOW if caller is using old sys_statfs(). */
1871 if (sizeof(long) < 8) {
1872 while (osfs.os_blocks > ~0UL && sfs->f_bsize < 0x40000000) {
1875 osfs.os_blocks >>= 1;
1876 osfs.os_bfree >>= 1;
1877 osfs.os_bavail >>= 1;
1881 sfs->f_blocks = osfs.os_blocks;
1882 sfs->f_bfree = osfs.os_bfree;
1883 sfs->f_bavail = osfs.os_bavail;
1884 sfs->f_fsid.val[0] = (__u32)fsid;
1885 sfs->f_fsid.val[1] = (__u32)(fsid >> 32);
1889 void ll_inode_size_lock(struct inode *inode)
1891 struct ll_inode_info *lli;
1893 LASSERT(!S_ISDIR(inode->i_mode));
1895 lli = ll_i2info(inode);
1896 mutex_lock(&lli->lli_size_mutex);
1899 void ll_inode_size_unlock(struct inode *inode)
1901 struct ll_inode_info *lli;
1903 lli = ll_i2info(inode);
1904 mutex_unlock(&lli->lli_size_mutex);
1907 void ll_update_inode_flags(struct inode *inode, int ext_flags)
1909 inode->i_flags = ll_ext_to_inode_flags(ext_flags);
1910 if (ext_flags & LUSTRE_PROJINHERIT_FL)
1911 ll_file_set_flag(ll_i2info(inode), LLIF_PROJECT_INHERIT);
1913 ll_file_clear_flag(ll_i2info(inode), LLIF_PROJECT_INHERIT);
1916 int ll_update_inode(struct inode *inode, struct lustre_md *md)
1918 struct ll_inode_info *lli = ll_i2info(inode);
1919 struct mdt_body *body = md->body;
1920 struct ll_sb_info *sbi = ll_i2sbi(inode);
1923 if (body->mbo_valid & OBD_MD_FLEASIZE) {
1924 rc = cl_file_inode_init(inode, md);
1929 if (S_ISDIR(inode->i_mode)) {
1930 rc = ll_update_lsm_md(inode, md);
1935 #ifdef CONFIG_FS_POSIX_ACL
1936 if (body->mbo_valid & OBD_MD_FLACL) {
1937 spin_lock(&lli->lli_lock);
1938 if (lli->lli_posix_acl)
1939 posix_acl_release(lli->lli_posix_acl);
1940 lli->lli_posix_acl = md->posix_acl;
1941 spin_unlock(&lli->lli_lock);
1944 inode->i_ino = cl_fid_build_ino(&body->mbo_fid1,
1945 sbi->ll_flags & LL_SBI_32BIT_API);
1946 inode->i_generation = cl_fid_build_gen(&body->mbo_fid1);
1948 if (body->mbo_valid & OBD_MD_FLATIME) {
1949 if (body->mbo_atime > inode->i_atime.tv_sec)
1950 inode->i_atime.tv_sec = body->mbo_atime;
1951 lli->lli_atime = body->mbo_atime;
1954 if (body->mbo_valid & OBD_MD_FLMTIME) {
1955 if (body->mbo_mtime > inode->i_mtime.tv_sec) {
1957 "setting ino %lu mtime from %lld to %llu\n",
1958 inode->i_ino, (s64)inode->i_mtime.tv_sec,
1960 inode->i_mtime.tv_sec = body->mbo_mtime;
1962 lli->lli_mtime = body->mbo_mtime;
1965 if (body->mbo_valid & OBD_MD_FLCTIME) {
1966 if (body->mbo_ctime > inode->i_ctime.tv_sec)
1967 inode->i_ctime.tv_sec = body->mbo_ctime;
1968 lli->lli_ctime = body->mbo_ctime;
1971 /* Clear i_flags to remove S_NOSEC before permissions are updated */
1972 if (body->mbo_valid & OBD_MD_FLFLAGS)
1973 ll_update_inode_flags(inode, body->mbo_flags);
1974 if (body->mbo_valid & OBD_MD_FLMODE)
1975 inode->i_mode = (inode->i_mode & S_IFMT) |
1976 (body->mbo_mode & ~S_IFMT);
1978 if (body->mbo_valid & OBD_MD_FLTYPE)
1979 inode->i_mode = (inode->i_mode & ~S_IFMT) |
1980 (body->mbo_mode & S_IFMT);
1982 LASSERT(inode->i_mode != 0);
1983 if (S_ISREG(inode->i_mode))
1984 inode->i_blkbits = min(PTLRPC_MAX_BRW_BITS + 1,
1985 LL_MAX_BLKSIZE_BITS);
1987 inode->i_blkbits = inode->i_sb->s_blocksize_bits;
1989 if (body->mbo_valid & OBD_MD_FLUID)
1990 inode->i_uid = make_kuid(&init_user_ns, body->mbo_uid);
1991 if (body->mbo_valid & OBD_MD_FLGID)
1992 inode->i_gid = make_kgid(&init_user_ns, body->mbo_gid);
1993 if (body->mbo_valid & OBD_MD_FLPROJID)
1994 lli->lli_projid = body->mbo_projid;
1995 if (body->mbo_valid & OBD_MD_FLNLINK)
1996 set_nlink(inode, body->mbo_nlink);
1997 if (body->mbo_valid & OBD_MD_FLRDEV)
1998 inode->i_rdev = old_decode_dev(body->mbo_rdev);
2000 if (body->mbo_valid & OBD_MD_FLID) {
2001 /* FID shouldn't be changed! */
2002 if (fid_is_sane(&lli->lli_fid)) {
2003 LASSERTF(lu_fid_eq(&lli->lli_fid, &body->mbo_fid1),
2004 "Trying to change FID "DFID
2005 " to the "DFID", inode "DFID"(%p)\n",
2006 PFID(&lli->lli_fid), PFID(&body->mbo_fid1),
2007 PFID(ll_inode2fid(inode)), inode);
2009 lli->lli_fid = body->mbo_fid1;
2013 LASSERT(fid_seq(&lli->lli_fid) != 0);
2015 if (body->mbo_valid & OBD_MD_FLSIZE) {
2016 i_size_write(inode, body->mbo_size);
2018 CDEBUG(D_VFSTRACE, "inode="DFID", updating i_size %llu\n",
2019 PFID(ll_inode2fid(inode)),
2020 (unsigned long long)body->mbo_size);
2022 if (body->mbo_valid & OBD_MD_FLBLOCKS)
2023 inode->i_blocks = body->mbo_blocks;
2026 if (body->mbo_valid & OBD_MD_TSTATE) {
2027 /* Set LLIF_FILE_RESTORING if restore ongoing and
2028 * clear it when done to ensure to start again
2029 * glimpsing updated attrs
2031 if (body->mbo_t_state & MS_RESTORE)
2032 ll_file_set_flag(lli, LLIF_FILE_RESTORING);
2034 ll_file_clear_flag(lli, LLIF_FILE_RESTORING);
2040 int ll_read_inode2(struct inode *inode, void *opaque)
2042 struct lustre_md *md = opaque;
2043 struct ll_inode_info *lli = ll_i2info(inode);
2047 CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p)\n",
2048 PFID(&lli->lli_fid), inode);
2050 /* Core attributes from the MDS first. This is a new inode, and
2051 * the VFS doesn't zero times in the core inode so we have to do
2052 * it ourselves. They will be overwritten by either MDS or OST
2053 * attributes - we just need to make sure they aren't newer.
2055 inode->i_mtime.tv_sec = 0;
2056 inode->i_atime.tv_sec = 0;
2057 inode->i_ctime.tv_sec = 0;
2059 rc = ll_update_inode(inode, md);
2063 /* OIDEBUG(inode); */
2065 #ifdef HAVE_BACKING_DEV_INFO
2066 /* initializing backing dev info. */
2067 inode->i_mapping->backing_dev_info = &s2lsi(inode->i_sb)->lsi_bdi;
2069 if (S_ISREG(inode->i_mode)) {
2070 struct ll_sb_info *sbi = ll_i2sbi(inode);
2071 inode->i_op = &ll_file_inode_operations;
2072 inode->i_fop = sbi->ll_fop;
2073 inode->i_mapping->a_ops = (struct address_space_operations *)&ll_aops;
2075 } else if (S_ISDIR(inode->i_mode)) {
2076 inode->i_op = &ll_dir_inode_operations;
2077 inode->i_fop = &ll_dir_operations;
2079 } else if (S_ISLNK(inode->i_mode)) {
2080 inode->i_op = &ll_fast_symlink_inode_operations;
2083 inode->i_op = &ll_special_inode_operations;
2085 init_special_inode(inode, inode->i_mode,
2094 void ll_delete_inode(struct inode *inode)
2096 struct ll_inode_info *lli = ll_i2info(inode);
2097 struct address_space *mapping = &inode->i_data;
2098 unsigned long nrpages;
2101 if (S_ISREG(inode->i_mode) && lli->lli_clob != NULL)
2102 /* It is last chance to write out dirty pages,
2103 * otherwise we may lose data while umount */
2104 cl_sync_file_range(inode, 0, OBD_OBJECT_EOF, CL_FSYNC_LOCAL, 1);
2106 truncate_inode_pages_final(mapping);
2108 /* Workaround for LU-118: Note nrpages may not be totally updated when
2109 * truncate_inode_pages() returns, as there can be a page in the process
2110 * of deletion (inside __delete_from_page_cache()) in the specified
2111 * range. Thus mapping->nrpages can be non-zero when this function
2112 * returns even after truncation of the whole mapping. Only do this if
2113 * npages isn't already zero.
2115 nrpages = mapping->nrpages;
2117 spin_lock_irq(&mapping->tree_lock);
2118 nrpages = mapping->nrpages;
2119 spin_unlock_irq(&mapping->tree_lock);
2120 } /* Workaround end */
2122 LASSERTF(nrpages == 0, "%s: inode="DFID"(%p) nrpages=%lu, "
2123 "see https://jira.whamcloud.com/browse/LU-118\n",
2124 ll_get_fsname(inode->i_sb, NULL, 0),
2125 PFID(ll_inode2fid(inode)), inode, nrpages);
2127 #ifdef HAVE_SBOPS_EVICT_INODE
2128 ll_clear_inode(inode);
2135 int ll_iocontrol(struct inode *inode, struct file *file,
2136 unsigned int cmd, unsigned long arg)
2138 struct ll_sb_info *sbi = ll_i2sbi(inode);
2139 struct ptlrpc_request *req = NULL;
2144 case FS_IOC_GETFLAGS: {
2145 struct mdt_body *body;
2146 struct md_op_data *op_data;
2148 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL,
2149 0, 0, LUSTRE_OPC_ANY,
2151 if (IS_ERR(op_data))
2152 RETURN(PTR_ERR(op_data));
2154 op_data->op_valid = OBD_MD_FLFLAGS;
2155 rc = md_getattr(sbi->ll_md_exp, op_data, &req);
2156 ll_finish_md_op_data(op_data);
2158 CERROR("%s: failure inode "DFID": rc = %d\n",
2159 sbi->ll_md_exp->exp_obd->obd_name,
2160 PFID(ll_inode2fid(inode)), rc);
2164 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
2166 flags = body->mbo_flags;
2168 ptlrpc_req_finished(req);
2170 RETURN(put_user(flags, (int __user *)arg));
2172 case FS_IOC_SETFLAGS: {
2174 struct md_op_data *op_data;
2175 struct cl_object *obj;
2176 struct fsxattr fa = { 0 };
2178 if (get_user(flags, (int __user *)arg))
2181 fa.fsx_projid = ll_i2info(inode)->lli_projid;
2182 if (flags & LUSTRE_PROJINHERIT_FL)
2183 fa.fsx_xflags = FS_XFLAG_PROJINHERIT;
2185 rc = ll_ioctl_check_project(inode, &fa);
2189 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
2190 LUSTRE_OPC_ANY, NULL);
2191 if (IS_ERR(op_data))
2192 RETURN(PTR_ERR(op_data));
2194 op_data->op_attr_flags = flags;
2195 op_data->op_xvalid |= OP_XVALID_FLAGS;
2196 rc = md_setattr(sbi->ll_md_exp, op_data, NULL, 0, &req);
2197 ll_finish_md_op_data(op_data);
2198 ptlrpc_req_finished(req);
2202 ll_update_inode_flags(inode, flags);
2204 obj = ll_i2info(inode)->lli_clob;
2208 OBD_ALLOC_PTR(attr);
2212 rc = cl_setattr_ost(obj, attr, OP_XVALID_FLAGS, flags);
2224 int ll_flush_ctx(struct inode *inode)
2226 struct ll_sb_info *sbi = ll_i2sbi(inode);
2228 CDEBUG(D_SEC, "flush context for user %d\n",
2229 from_kuid(&init_user_ns, current_uid()));
2231 obd_set_info_async(NULL, sbi->ll_md_exp,
2232 sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
2234 obd_set_info_async(NULL, sbi->ll_dt_exp,
2235 sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
2240 /* umount -f client means force down, don't save state */
2241 void ll_umount_begin(struct super_block *sb)
2243 struct ll_sb_info *sbi = ll_s2sbi(sb);
2244 struct obd_device *obd;
2245 struct obd_ioctl_data *ioc_data;
2246 struct l_wait_info lwi;
2247 wait_queue_head_t waitq;
2250 CDEBUG(D_VFSTRACE, "VFS Op: superblock %p count %d active %d\n", sb,
2251 sb->s_count, atomic_read(&sb->s_active));
2253 obd = class_exp2obd(sbi->ll_md_exp);
2255 CERROR("Invalid MDC connection handle %#llx\n",
2256 sbi->ll_md_exp->exp_handle.h_cookie);
2262 obd = class_exp2obd(sbi->ll_dt_exp);
2264 CERROR("Invalid LOV connection handle %#llx\n",
2265 sbi->ll_dt_exp->exp_handle.h_cookie);
2271 OBD_ALLOC_PTR(ioc_data);
2273 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_md_exp,
2274 sizeof *ioc_data, ioc_data, NULL);
2276 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_dt_exp,
2277 sizeof *ioc_data, ioc_data, NULL);
2279 OBD_FREE_PTR(ioc_data);
2282 /* Really, we'd like to wait until there are no requests outstanding,
2283 * and then continue. For now, we just periodically checking for vfs
2284 * to decrement mnt_cnt and hope to finish it within 10sec.
2286 init_waitqueue_head(&waitq);
2287 lwi = LWI_TIMEOUT_INTERVAL(cfs_time_seconds(10),
2288 cfs_time_seconds(1), NULL, NULL);
2289 l_wait_event(waitq, may_umount(sbi->ll_mnt.mnt), &lwi);
2294 int ll_remount_fs(struct super_block *sb, int *flags, char *data)
2296 struct ll_sb_info *sbi = ll_s2sbi(sb);
2297 char *profilenm = get_profile_name(sb);
2301 if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY)) {
2302 read_only = *flags & MS_RDONLY;
2303 err = obd_set_info_async(NULL, sbi->ll_md_exp,
2304 sizeof(KEY_READ_ONLY),
2305 KEY_READ_ONLY, sizeof(read_only),
2308 LCONSOLE_WARN("Failed to remount %s %s (%d)\n",
2309 profilenm, read_only ?
2310 "read-only" : "read-write", err);
2315 sb->s_flags |= MS_RDONLY;
2317 sb->s_flags &= ~MS_RDONLY;
2319 if (sbi->ll_flags & LL_SBI_VERBOSE)
2320 LCONSOLE_WARN("Remounted %s %s\n", profilenm,
2321 read_only ? "read-only" : "read-write");
2327 * Cleanup the open handle that is cached on MDT-side.
2329 * For open case, the client side open handling thread may hit error
2330 * after the MDT grant the open. Under such case, the client should
2331 * send close RPC to the MDT as cleanup; otherwise, the open handle
2332 * on the MDT will be leaked there until the client umount or evicted.
2334 * In further, if someone unlinked the file, because the open handle
2335 * holds the reference on such file/object, then it will block the
2336 * subsequent threads that want to locate such object via FID.
2338 * \param[in] sb super block for this file-system
2339 * \param[in] open_req pointer to the original open request
2341 void ll_open_cleanup(struct super_block *sb, struct ptlrpc_request *open_req)
2343 struct mdt_body *body;
2344 struct md_op_data *op_data;
2345 struct ptlrpc_request *close_req = NULL;
2346 struct obd_export *exp = ll_s2sbi(sb)->ll_md_exp;
2349 body = req_capsule_server_get(&open_req->rq_pill, &RMF_MDT_BODY);
2350 OBD_ALLOC_PTR(op_data);
2351 if (op_data == NULL) {
2352 CWARN("%s: cannot allocate op_data to release open handle for "
2354 ll_get_fsname(sb, NULL, 0), PFID(&body->mbo_fid1));
2359 op_data->op_fid1 = body->mbo_fid1;
2360 op_data->op_open_handle = body->mbo_open_handle;
2361 op_data->op_mod_time = ktime_get_real_seconds();
2362 md_close(exp, op_data, NULL, &close_req);
2363 ptlrpc_req_finished(close_req);
2364 ll_finish_md_op_data(op_data);
2369 int ll_prep_inode(struct inode **inode, struct ptlrpc_request *req,
2370 struct super_block *sb, struct lookup_intent *it)
2372 struct ll_sb_info *sbi = NULL;
2373 struct lustre_md md = { NULL };
2377 LASSERT(*inode || sb);
2378 sbi = sb ? ll_s2sbi(sb) : ll_i2sbi(*inode);
2379 rc = md_get_lustre_md(sbi->ll_md_exp, req, sbi->ll_dt_exp,
2380 sbi->ll_md_exp, &md);
2385 rc = ll_update_inode(*inode, &md);
2389 LASSERT(sb != NULL);
2392 * At this point server returns to client's same fid as client
2393 * generated for creating. So using ->fid1 is okay here.
2395 if (!fid_is_sane(&md.body->mbo_fid1)) {
2396 CERROR("%s: Fid is insane "DFID"\n",
2397 ll_get_fsname(sb, NULL, 0),
2398 PFID(&md.body->mbo_fid1));
2399 GOTO(out, rc = -EINVAL);
2402 *inode = ll_iget(sb, cl_fid_build_ino(&md.body->mbo_fid1,
2403 sbi->ll_flags & LL_SBI_32BIT_API),
2405 if (IS_ERR(*inode)) {
2406 #ifdef CONFIG_FS_POSIX_ACL
2408 posix_acl_release(md.posix_acl);
2409 md.posix_acl = NULL;
2412 rc = IS_ERR(*inode) ? PTR_ERR(*inode) : -ENOMEM;
2414 CERROR("new_inode -fatal: rc %d\n", rc);
2419 /* Handling piggyback layout lock.
2420 * Layout lock can be piggybacked by getattr and open request.
2421 * The lsm can be applied to inode only if it comes with a layout lock
2422 * otherwise correct layout may be overwritten, for example:
2423 * 1. proc1: mdt returns a lsm but not granting layout
2424 * 2. layout was changed by another client
2425 * 3. proc2: refresh layout and layout lock granted
2426 * 4. proc1: to apply a stale layout */
2427 if (it != NULL && it->it_lock_mode != 0) {
2428 struct lustre_handle lockh;
2429 struct ldlm_lock *lock;
2431 lockh.cookie = it->it_lock_handle;
2432 lock = ldlm_handle2lock(&lockh);
2433 LASSERT(lock != NULL);
2434 if (ldlm_has_layout(lock)) {
2435 struct cl_object_conf conf;
2437 memset(&conf, 0, sizeof(conf));
2438 conf.coc_opc = OBJECT_CONF_SET;
2439 conf.coc_inode = *inode;
2440 conf.coc_lock = lock;
2441 conf.u.coc_layout = md.layout;
2442 (void)ll_layout_conf(*inode, &conf);
2444 LDLM_LOCK_PUT(lock);
2450 md_free_lustre_md(sbi->ll_md_exp, &md);
2453 if (rc != 0 && it != NULL && it->it_op & IT_OPEN)
2454 ll_open_cleanup(sb != NULL ? sb : (*inode)->i_sb, req);
2459 int ll_obd_statfs(struct inode *inode, void __user *arg)
2461 struct ll_sb_info *sbi = NULL;
2462 struct obd_export *exp;
2464 struct obd_ioctl_data *data = NULL;
2468 if (!inode || !(sbi = ll_i2sbi(inode)))
2469 GOTO(out_statfs, rc = -EINVAL);
2471 rc = obd_ioctl_getdata(&buf, &len, arg);
2473 GOTO(out_statfs, rc);
2476 if (!data->ioc_inlbuf1 || !data->ioc_inlbuf2 ||
2477 !data->ioc_pbuf1 || !data->ioc_pbuf2)
2478 GOTO(out_statfs, rc = -EINVAL);
2480 if (data->ioc_inllen1 != sizeof(__u32) ||
2481 data->ioc_inllen2 != sizeof(__u32) ||
2482 data->ioc_plen1 != sizeof(struct obd_statfs) ||
2483 data->ioc_plen2 != sizeof(struct obd_uuid))
2484 GOTO(out_statfs, rc = -EINVAL);
2486 memcpy(&type, data->ioc_inlbuf1, sizeof(__u32));
2487 if (type & LL_STATFS_LMV)
2488 exp = sbi->ll_md_exp;
2489 else if (type & LL_STATFS_LOV)
2490 exp = sbi->ll_dt_exp;
2492 GOTO(out_statfs, rc = -ENODEV);
2494 rc = obd_iocontrol(IOC_OBD_STATFS, exp, len, buf, NULL);
2496 GOTO(out_statfs, rc);
2498 OBD_FREE_LARGE(buf, len);
2503 * this is normally called in ll_fini_md_op_data(), but sometimes it needs to
2504 * be called early to avoid deadlock.
2506 void ll_unlock_md_op_lsm(struct md_op_data *op_data)
2508 if (op_data->op_mea2_sem) {
2509 up_read(op_data->op_mea2_sem);
2510 op_data->op_mea2_sem = NULL;
2513 if (op_data->op_mea1_sem) {
2514 up_read(op_data->op_mea1_sem);
2515 op_data->op_mea1_sem = NULL;
2519 /* this function prepares md_op_data hint for passing it down to MD stack. */
2520 struct md_op_data *ll_prep_md_op_data(struct md_op_data *op_data,
2521 struct inode *i1, struct inode *i2,
2522 const char *name, size_t namelen,
2523 __u32 mode, __u32 opc, void *data)
2525 LASSERT(i1 != NULL);
2528 /* Do not reuse namelen for something else. */
2530 return ERR_PTR(-EINVAL);
2532 if (namelen > ll_i2sbi(i1)->ll_namelen)
2533 return ERR_PTR(-ENAMETOOLONG);
2535 if (!lu_name_is_valid_2(name, namelen))
2536 return ERR_PTR(-EINVAL);
2539 if (op_data == NULL)
2540 OBD_ALLOC_PTR(op_data);
2542 if (op_data == NULL)
2543 return ERR_PTR(-ENOMEM);
2545 ll_i2gids(op_data->op_suppgids, i1, i2);
2546 op_data->op_fid1 = *ll_inode2fid(i1);
2547 op_data->op_default_stripe_offset = -1;
2549 if (S_ISDIR(i1->i_mode)) {
2550 down_read(&ll_i2info(i1)->lli_lsm_sem);
2551 op_data->op_mea1_sem = &ll_i2info(i1)->lli_lsm_sem;
2552 op_data->op_mea1 = ll_i2info(i1)->lli_lsm_md;
2553 if (opc == LUSTRE_OPC_MKDIR)
2554 op_data->op_default_stripe_offset =
2555 ll_i2info(i1)->lli_def_stripe_offset;
2559 op_data->op_fid2 = *ll_inode2fid(i2);
2560 if (S_ISDIR(i2->i_mode)) {
2562 down_read(&ll_i2info(i2)->lli_lsm_sem);
2563 op_data->op_mea2_sem =
2564 &ll_i2info(i2)->lli_lsm_sem;
2566 op_data->op_mea2 = ll_i2info(i2)->lli_lsm_md;
2569 fid_zero(&op_data->op_fid2);
2572 if (ll_i2sbi(i1)->ll_flags & LL_SBI_64BIT_HASH)
2573 op_data->op_cli_flags |= CLI_HASH64;
2575 if (ll_need_32bit_api(ll_i2sbi(i1)))
2576 op_data->op_cli_flags |= CLI_API32;
2578 op_data->op_name = name;
2579 op_data->op_namelen = namelen;
2580 op_data->op_mode = mode;
2581 op_data->op_mod_time = ktime_get_real_seconds();
2582 op_data->op_fsuid = from_kuid(&init_user_ns, current_fsuid());
2583 op_data->op_fsgid = from_kgid(&init_user_ns, current_fsgid());
2584 op_data->op_cap = cfs_curproc_cap_pack();
2585 op_data->op_mds = 0;
2586 if ((opc == LUSTRE_OPC_CREATE) && (name != NULL) &&
2587 filename_is_volatile(name, namelen, &op_data->op_mds)) {
2588 op_data->op_bias |= MDS_CREATE_VOLATILE;
2590 op_data->op_data = data;
2595 void ll_finish_md_op_data(struct md_op_data *op_data)
2597 ll_unlock_md_op_lsm(op_data);
2598 security_release_secctx(op_data->op_file_secctx,
2599 op_data->op_file_secctx_size);
2600 OBD_FREE_PTR(op_data);
2603 #ifdef HAVE_SUPEROPS_USE_DENTRY
2604 int ll_show_options(struct seq_file *seq, struct dentry *dentry)
2606 int ll_show_options(struct seq_file *seq, struct vfsmount *vfs)
2609 struct ll_sb_info *sbi;
2611 #ifdef HAVE_SUPEROPS_USE_DENTRY
2612 LASSERT((seq != NULL) && (dentry != NULL));
2613 sbi = ll_s2sbi(dentry->d_sb);
2615 LASSERT((seq != NULL) && (vfs != NULL));
2616 sbi = ll_s2sbi(vfs->mnt_sb);
2619 if (sbi->ll_flags & LL_SBI_NOLCK)
2620 seq_puts(seq, ",nolock");
2622 if (sbi->ll_flags & LL_SBI_FLOCK)
2623 seq_puts(seq, ",flock");
2625 if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
2626 seq_puts(seq, ",localflock");
2628 if (sbi->ll_flags & LL_SBI_USER_XATTR)
2629 seq_puts(seq, ",user_xattr");
2631 if (sbi->ll_flags & LL_SBI_LAZYSTATFS)
2632 seq_puts(seq, ",lazystatfs");
2634 if (sbi->ll_flags & LL_SBI_USER_FID2PATH)
2635 seq_puts(seq, ",user_fid2path");
2637 if (sbi->ll_flags & LL_SBI_ALWAYS_PING)
2638 seq_puts(seq, ",always_ping");
2644 * Get obd name by cmd, and copy out to user space
2646 int ll_get_obd_name(struct inode *inode, unsigned int cmd, unsigned long arg)
2648 struct ll_sb_info *sbi = ll_i2sbi(inode);
2649 struct obd_device *obd;
2652 if (cmd == OBD_IOC_GETDTNAME)
2653 obd = class_exp2obd(sbi->ll_dt_exp);
2654 else if (cmd == OBD_IOC_GETMDNAME)
2655 obd = class_exp2obd(sbi->ll_md_exp);
2662 if (copy_to_user((void __user *)arg, obd->obd_name,
2663 strlen(obd->obd_name) + 1))
2670 * Get lustre file system name by \a sbi. If \a buf is provided(non-NULL), the
2671 * fsname will be returned in this buffer; otherwise, a static buffer will be
2672 * used to store the fsname and returned to caller.
2674 char *ll_get_fsname(struct super_block *sb, char *buf, int buflen)
2676 static char fsname_static[MTI_NAME_MAXLEN];
2677 struct lustre_sb_info *lsi = s2lsi(sb);
2682 /* this means the caller wants to use static buffer
2683 * and it doesn't care about race. Usually this is
2684 * in error reporting path */
2685 buf = fsname_static;
2686 buflen = sizeof(fsname_static);
2689 len = strlen(lsi->lsi_lmd->lmd_profile);
2690 ptr = strrchr(lsi->lsi_lmd->lmd_profile, '-');
2691 if (ptr && (strcmp(ptr, "-client") == 0))
2694 if (unlikely(len >= buflen))
2696 strncpy(buf, lsi->lsi_lmd->lmd_profile, len);
2702 static char* ll_d_path(struct dentry *dentry, char *buf, int bufsize)
2709 p.mnt = current->fs->root.mnt;
2711 path = d_path(&p, buf, bufsize);
2716 void ll_dirty_page_discard_warn(struct page *page, int ioret)
2718 char *buf, *path = NULL;
2719 struct dentry *dentry = NULL;
2720 struct inode *inode = page->mapping->host;
2722 /* this can be called inside spin lock so use GFP_ATOMIC. */
2723 buf = (char *)__get_free_page(GFP_ATOMIC);
2725 dentry = d_find_alias(page->mapping->host);
2727 path = ll_d_path(dentry, buf, PAGE_SIZE);
2731 "%s: dirty page discard: %s/fid: "DFID"/%s may get corrupted "
2732 "(rc %d)\n", ll_get_fsname(page->mapping->host->i_sb, NULL, 0),
2733 s2lsi(page->mapping->host->i_sb)->lsi_lmd->lmd_dev,
2734 PFID(ll_inode2fid(inode)),
2735 (path && !IS_ERR(path)) ? path : "", ioret);
2741 free_page((unsigned long)buf);
2744 ssize_t ll_copy_user_md(const struct lov_user_md __user *md,
2745 struct lov_user_md **kbuf)
2747 struct lov_user_md lum;
2751 if (copy_from_user(&lum, md, sizeof(lum)))
2754 lum_size = ll_lov_user_md_size(&lum);
2758 OBD_ALLOC(*kbuf, lum_size);
2762 if (copy_from_user(*kbuf, md, lum_size) != 0) {
2763 OBD_FREE(*kbuf, lum_size);
2771 * Compute llite root squash state after a change of root squash
2772 * configuration setting or add/remove of a lnet nid
2774 void ll_compute_rootsquash_state(struct ll_sb_info *sbi)
2776 struct root_squash_info *squash = &sbi->ll_squash;
2779 struct lnet_process_id id;
2781 /* Update norootsquash flag */
2782 down_write(&squash->rsi_sem);
2783 if (list_empty(&squash->rsi_nosquash_nids))
2784 sbi->ll_flags &= ~LL_SBI_NOROOTSQUASH;
2786 /* Do not apply root squash as soon as one of our NIDs is
2787 * in the nosquash_nids list */
2790 while (LNetGetId(i++, &id) != -ENOENT) {
2791 if (LNET_NETTYP(LNET_NIDNET(id.nid)) == LOLND)
2793 if (cfs_match_nid(id.nid, &squash->rsi_nosquash_nids)) {
2799 sbi->ll_flags |= LL_SBI_NOROOTSQUASH;
2801 sbi->ll_flags &= ~LL_SBI_NOROOTSQUASH;
2803 up_write(&squash->rsi_sem);
2807 * Parse linkea content to extract information about a given hardlink
2809 * \param[in] ldata - Initialized linkea data
2810 * \param[in] linkno - Link identifier
2811 * \param[out] parent_fid - The entry's parent FID
2812 * \param[out] ln - Entry name destination buffer
2814 * \retval 0 on success
2815 * \retval Appropriate negative error code on failure
2817 static int ll_linkea_decode(struct linkea_data *ldata, unsigned int linkno,
2818 struct lu_fid *parent_fid, struct lu_name *ln)
2824 rc = linkea_init_with_rec(ldata);
2828 if (linkno >= ldata->ld_leh->leh_reccount)
2829 /* beyond last link */
2832 linkea_first_entry(ldata);
2833 for (idx = 0; ldata->ld_lee != NULL; idx++) {
2834 linkea_entry_unpack(ldata->ld_lee, &ldata->ld_reclen, ln,
2839 linkea_next_entry(ldata);
2849 * Get parent FID and name of an identified link. Operation is performed for
2850 * a given link number, letting the caller iterate over linkno to list one or
2851 * all links of an entry.
2853 * \param[in] file - File descriptor against which to perform the operation
2854 * \param[in,out] arg - User-filled structure containing the linkno to operate
2855 * on and the available size. It is eventually filled with
2856 * the requested information or left untouched on error
2858 * \retval - 0 on success
2859 * \retval - Appropriate negative error code on failure
2861 int ll_getparent(struct file *file, struct getparent __user *arg)
2863 struct inode *inode = file_inode(file);
2864 struct linkea_data *ldata;
2865 struct lu_buf buf = LU_BUF_NULL;
2867 struct lu_fid parent_fid;
2874 if (!cfs_capable(CFS_CAP_DAC_READ_SEARCH) &&
2875 !(ll_i2sbi(inode)->ll_flags & LL_SBI_USER_FID2PATH))
2878 if (get_user(name_size, &arg->gp_name_size))
2881 if (get_user(linkno, &arg->gp_linkno))
2884 if (name_size > PATH_MAX)
2887 OBD_ALLOC(ldata, sizeof(*ldata));
2891 rc = linkea_data_new(ldata, &buf);
2893 GOTO(ldata_free, rc);
2895 #ifdef HAVE_XATTR_HANDLER_FLAGS
2896 rc = ll_xattr_list(inode, XATTR_NAME_LINK, XATTR_TRUSTED_T, buf.lb_buf,
2897 buf.lb_len, OBD_MD_FLXATTR);
2899 rc = ll_getxattr(file_dentry(file), XATTR_NAME_LINK, buf.lb_buf,
2901 #endif /* HAVE_XATTR_HANDLER_FLAGS */
2905 rc = ll_linkea_decode(ldata, linkno, &parent_fid, &ln);
2909 if (ln.ln_namelen >= name_size)
2910 GOTO(lb_free, rc = -EOVERFLOW);
2912 if (copy_to_user(&arg->gp_fid, &parent_fid, sizeof(arg->gp_fid)))
2913 GOTO(lb_free, rc = -EFAULT);
2915 if (copy_to_user(&arg->gp_name, ln.ln_name, ln.ln_namelen))
2916 GOTO(lb_free, rc = -EFAULT);
2918 if (put_user('\0', arg->gp_name + ln.ln_namelen))
2919 GOTO(lb_free, rc = -EFAULT);
2924 OBD_FREE(ldata, sizeof(*ldata));