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, 2016, 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_ioctl.h>
52 #include <lustre_ha.h>
53 #include <lustre_dlm.h>
54 #include <lprocfs_status.h>
55 #include <lustre_disk.h>
56 #include <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 =
103 SBI_DEFAULT_READAHEAD_WHOLE_MAX;
105 ll_generate_random_uuid(uuid);
106 class_uuid_unparse(uuid, &sbi->ll_sb_uuid);
107 CDEBUG(D_CONFIG, "generated uuid: %s\n", sbi->ll_sb_uuid.uuid);
109 sbi->ll_flags |= LL_SBI_VERBOSE;
110 #ifdef ENABLE_CHECKSUM
111 sbi->ll_flags |= LL_SBI_CHECKSUM;
114 #ifdef HAVE_LRU_RESIZE_SUPPORT
115 sbi->ll_flags |= LL_SBI_LRU_RESIZE;
117 sbi->ll_flags |= LL_SBI_LAZYSTATFS;
119 for (i = 0; i <= LL_PROCESS_HIST_MAX; i++) {
120 spin_lock_init(&sbi->ll_rw_extents_info.pp_extents[i].
122 spin_lock_init(&sbi->ll_rw_extents_info.pp_extents[i].
126 /* metadata statahead is enabled by default */
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;
136 sbi->ll_squash.rsi_uid = 0;
137 sbi->ll_squash.rsi_gid = 0;
138 INIT_LIST_HEAD(&sbi->ll_squash.rsi_nosquash_nids);
139 init_rwsem(&sbi->ll_squash.rsi_sem);
144 static void ll_free_sbi(struct super_block *sb)
146 struct ll_sb_info *sbi = ll_s2sbi(sb);
150 if (!list_empty(&sbi->ll_squash.rsi_nosquash_nids))
151 cfs_free_nidlist(&sbi->ll_squash.rsi_nosquash_nids);
152 if (sbi->ll_cache != NULL) {
153 cl_cache_decref(sbi->ll_cache);
154 sbi->ll_cache = NULL;
156 OBD_FREE(sbi, sizeof(*sbi));
161 static inline int obd_connect_has_secctx(struct obd_connect_data *data)
163 return data->ocd_connect_flags & OBD_CONNECT_FLAGS2 &&
164 data->ocd_connect_flags2 & OBD_CONNECT2_FILE_SECCTX;
167 static int client_common_fill_super(struct super_block *sb, char *md, char *dt,
168 struct vfsmount *mnt)
170 struct inode *root = NULL;
171 struct ll_sb_info *sbi = ll_s2sbi(sb);
172 struct obd_device *obd;
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 obd = class_name2obd(md);
185 CERROR("MD %s: not setup or attached\n", md);
199 /* indicate the features supported by this client */
200 data->ocd_connect_flags = OBD_CONNECT_IBITS | OBD_CONNECT_NODEVOH |
201 OBD_CONNECT_ATTRFID |
202 OBD_CONNECT_VERSION | OBD_CONNECT_BRW_SIZE |
203 OBD_CONNECT_MDS_CAPA | OBD_CONNECT_OSS_CAPA |
204 OBD_CONNECT_CANCELSET | OBD_CONNECT_FID |
205 OBD_CONNECT_AT | OBD_CONNECT_LOV_V3 |
206 OBD_CONNECT_VBR | OBD_CONNECT_FULL20 |
207 OBD_CONNECT_64BITHASH |
208 OBD_CONNECT_EINPROGRESS |
209 OBD_CONNECT_JOBSTATS | OBD_CONNECT_LVB_TYPE |
210 OBD_CONNECT_LAYOUTLOCK | OBD_CONNECT_PINGLESS|
211 OBD_CONNECT_MAX_EASIZE |
212 OBD_CONNECT_FLOCK_DEAD |
213 OBD_CONNECT_DISP_STRIPE | OBD_CONNECT_LFSCK |
214 OBD_CONNECT_OPEN_BY_FID |
215 OBD_CONNECT_DIR_STRIPE |
216 OBD_CONNECT_BULK_MBITS |
217 OBD_CONNECT_SUBTREE |
218 OBD_CONNECT_FLAGS2 | OBD_CONNECT_MULTIMODRPCS;
220 data->ocd_connect_flags2 = 0;
222 #ifdef HAVE_LRU_RESIZE_SUPPORT
223 if (sbi->ll_flags & LL_SBI_LRU_RESIZE)
224 data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
226 #ifdef CONFIG_FS_POSIX_ACL
227 data->ocd_connect_flags |= OBD_CONNECT_ACL | OBD_CONNECT_UMASK;
230 if (OBD_FAIL_CHECK(OBD_FAIL_MDC_LIGHTWEIGHT))
231 /* flag mdc connection as lightweight, only used for test
232 * purpose, use with care */
233 data->ocd_connect_flags |= OBD_CONNECT_LIGHTWEIGHT;
235 data->ocd_ibits_known = MDS_INODELOCK_FULL;
236 data->ocd_version = LUSTRE_VERSION_CODE;
238 if (sb->s_flags & MS_RDONLY)
239 data->ocd_connect_flags |= OBD_CONNECT_RDONLY;
240 if (sbi->ll_flags & LL_SBI_USER_XATTR)
241 data->ocd_connect_flags |= OBD_CONNECT_XATTR;
243 if (sbi->ll_flags & LL_SBI_FLOCK)
244 sbi->ll_fop = &ll_file_operations_flock;
245 else if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
246 sbi->ll_fop = &ll_file_operations;
248 sbi->ll_fop = &ll_file_operations_noflock;
250 /* always ping even if server suppress_pings */
251 if (sbi->ll_flags & LL_SBI_ALWAYS_PING)
252 data->ocd_connect_flags &= ~OBD_CONNECT_PINGLESS;
254 #ifdef HAVE_SECURITY_DENTRY_INIT_SECURITY
255 data->ocd_connect_flags2 |= OBD_CONNECT2_FILE_SECCTX;
256 #endif /* HAVE_SECURITY_DENTRY_INIT_SECURITY */
258 data->ocd_brw_size = MD_MAX_BRW_SIZE;
260 err = obd_connect(NULL, &sbi->ll_md_exp, obd, &sbi->ll_sb_uuid, data, NULL);
262 LCONSOLE_ERROR_MSG(0x14f, "An MDT (md %s) is performing "
263 "recovery, of which this client is not a "
264 "part. Please wait for recovery to complete,"
265 " abort, or time out.\n", md);
268 CERROR("cannot connect to %s: rc = %d\n", md, err);
272 sbi->ll_md_exp->exp_connect_data = *data;
274 err = obd_fid_init(sbi->ll_md_exp->exp_obd, sbi->ll_md_exp,
275 LUSTRE_SEQ_METADATA);
277 CERROR("%s: Can't init metadata layer FID infrastructure, "
278 "rc = %d\n", sbi->ll_md_exp->exp_obd->obd_name, err);
282 /* For mount, we only need fs info from MDT0, and also in DNE, it
283 * can make sure the client can be mounted as long as MDT0 is
285 err = obd_statfs(NULL, sbi->ll_md_exp, osfs,
286 cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS),
287 OBD_STATFS_FOR_MDT0);
289 GOTO(out_md_fid, err);
291 /* This needs to be after statfs to ensure connect has finished.
292 * Note that "data" does NOT contain the valid connect reply.
293 * If connecting to a 1.8 server there will be no LMV device, so
294 * we can access the MDC export directly and exp_connect_flags will
295 * be non-zero, but if accessing an upgraded 2.1 server it will
296 * have the correct flags filled in.
297 * XXX: fill in the LMV exp_connect_flags from MDC(s). */
298 valid = exp_connect_flags(sbi->ll_md_exp) & CLIENT_CONNECT_MDT_REQD;
299 if (exp_connect_flags(sbi->ll_md_exp) != 0 &&
300 valid != CLIENT_CONNECT_MDT_REQD) {
303 OBD_ALLOC_WAIT(buf, PAGE_SIZE);
304 obd_connect_flags2str(buf, PAGE_SIZE,
305 valid ^ CLIENT_CONNECT_MDT_REQD, 0, ",");
306 LCONSOLE_ERROR_MSG(0x170, "Server %s does not support "
307 "feature(s) needed for correct operation "
308 "of this client (%s). Please upgrade "
309 "server or downgrade client.\n",
310 sbi->ll_md_exp->exp_obd->obd_name, buf);
311 OBD_FREE(buf, PAGE_SIZE);
312 GOTO(out_md_fid, err = -EPROTO);
315 size = sizeof(*data);
316 err = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_CONN_DATA),
317 KEY_CONN_DATA, &size, data);
319 CERROR("%s: Get connect data failed: rc = %d\n",
320 sbi->ll_md_exp->exp_obd->obd_name, err);
321 GOTO(out_md_fid, err);
324 LASSERT(osfs->os_bsize);
325 sb->s_blocksize = osfs->os_bsize;
326 sb->s_blocksize_bits = log2(osfs->os_bsize);
327 sb->s_magic = LL_SUPER_MAGIC;
328 sb->s_maxbytes = MAX_LFS_FILESIZE;
329 sbi->ll_namelen = osfs->os_namelen;
330 sbi->ll_mnt.mnt = current->fs->root.mnt;
332 if ((sbi->ll_flags & LL_SBI_USER_XATTR) &&
333 !(data->ocd_connect_flags & OBD_CONNECT_XATTR)) {
334 LCONSOLE_INFO("Disabling user_xattr feature because "
335 "it is not supported on the server\n");
336 sbi->ll_flags &= ~LL_SBI_USER_XATTR;
339 if (data->ocd_connect_flags & OBD_CONNECT_ACL) {
341 sb->s_flags |= MS_POSIXACL;
343 sbi->ll_flags |= LL_SBI_ACL;
345 LCONSOLE_INFO("client wants to enable acl, but mdt not!\n");
347 sb->s_flags &= ~MS_POSIXACL;
349 sbi->ll_flags &= ~LL_SBI_ACL;
352 if (data->ocd_connect_flags & OBD_CONNECT_64BITHASH)
353 sbi->ll_flags |= LL_SBI_64BIT_HASH;
355 if (data->ocd_connect_flags & OBD_CONNECT_LAYOUTLOCK)
356 sbi->ll_flags |= LL_SBI_LAYOUT_LOCK;
358 if (obd_connect_has_secctx(data))
359 sbi->ll_flags |= LL_SBI_FILE_SECCTX;
361 if (data->ocd_ibits_known & MDS_INODELOCK_XATTR) {
362 if (!(data->ocd_connect_flags & OBD_CONNECT_MAX_EASIZE)) {
363 LCONSOLE_INFO("%s: disabling xattr cache due to "
364 "unknown maximum xattr size.\n", dt);
366 sbi->ll_flags |= LL_SBI_XATTR_CACHE;
367 sbi->ll_xattr_cache_enabled = 1;
371 obd = class_name2obd(dt);
373 CERROR("DT %s: not setup or attached\n", dt);
374 GOTO(out_md_fid, err = -ENODEV);
377 /* pass client page size via ocd_grant_blkbits, the server should report
378 * back its backend blocksize for grant calculation purpose */
379 data->ocd_grant_blkbits = PAGE_SHIFT;
381 data->ocd_connect_flags = OBD_CONNECT_GRANT | OBD_CONNECT_VERSION |
382 OBD_CONNECT_REQPORTAL | OBD_CONNECT_BRW_SIZE |
383 OBD_CONNECT_CANCELSET | OBD_CONNECT_FID |
384 OBD_CONNECT_SRVLOCK | OBD_CONNECT_TRUNCLOCK|
385 OBD_CONNECT_AT | OBD_CONNECT_OSS_CAPA |
386 OBD_CONNECT_VBR | OBD_CONNECT_FULL20 |
387 OBD_CONNECT_64BITHASH | OBD_CONNECT_MAXBYTES |
388 OBD_CONNECT_EINPROGRESS |
389 OBD_CONNECT_JOBSTATS | OBD_CONNECT_LVB_TYPE |
390 OBD_CONNECT_LAYOUTLOCK |
391 OBD_CONNECT_PINGLESS | OBD_CONNECT_LFSCK |
392 OBD_CONNECT_BULK_MBITS;
394 data->ocd_connect_flags2 = 0;
396 if (!OBD_FAIL_CHECK(OBD_FAIL_OSC_CONNECT_GRANT_PARAM))
397 data->ocd_connect_flags |= OBD_CONNECT_GRANT_PARAM;
399 /* OBD_CONNECT_CKSUM should always be set, even if checksums are
400 * disabled by default, because it can still be enabled on the
401 * fly via /proc. As a consequence, we still need to come to an
402 * agreement on the supported algorithms at connect time */
403 data->ocd_connect_flags |= OBD_CONNECT_CKSUM;
405 if (OBD_FAIL_CHECK(OBD_FAIL_OSC_CKSUM_ADLER_ONLY))
406 data->ocd_cksum_types = OBD_CKSUM_ADLER;
408 data->ocd_cksum_types = cksum_types_supported_client();
410 #ifdef HAVE_LRU_RESIZE_SUPPORT
411 data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
413 /* always ping even if server suppress_pings */
414 if (sbi->ll_flags & LL_SBI_ALWAYS_PING)
415 data->ocd_connect_flags &= ~OBD_CONNECT_PINGLESS;
417 CDEBUG(D_RPCTRACE, "ocd_connect_flags: %#llx ocd_version: %d "
418 "ocd_grant: %d\n", data->ocd_connect_flags,
419 data->ocd_version, data->ocd_grant);
421 obd->obd_upcall.onu_owner = &sbi->ll_lco;
422 obd->obd_upcall.onu_upcall = cl_ocd_update;
424 data->ocd_brw_size = DT_MAX_BRW_SIZE;
426 err = obd_connect(NULL, &sbi->ll_dt_exp, obd, &sbi->ll_sb_uuid, data,
429 LCONSOLE_ERROR_MSG(0x150, "An OST (dt %s) is performing "
430 "recovery, of which this client is not a "
431 "part. Please wait for recovery to "
432 "complete, abort, or time out.\n", dt);
435 CERROR("%s: Cannot connect to %s: rc = %d\n",
436 sbi->ll_dt_exp->exp_obd->obd_name, dt, err);
440 sbi->ll_dt_exp->exp_connect_data = *data;
442 err = obd_fid_init(sbi->ll_dt_exp->exp_obd, sbi->ll_dt_exp,
443 LUSTRE_SEQ_METADATA);
445 CERROR("%s: Can't init data layer FID infrastructure, "
446 "rc = %d\n", sbi->ll_dt_exp->exp_obd->obd_name, err);
450 mutex_lock(&sbi->ll_lco.lco_lock);
451 sbi->ll_lco.lco_flags = data->ocd_connect_flags;
452 sbi->ll_lco.lco_md_exp = sbi->ll_md_exp;
453 sbi->ll_lco.lco_dt_exp = sbi->ll_dt_exp;
454 mutex_unlock(&sbi->ll_lco.lco_lock);
456 fid_zero(&sbi->ll_root_fid);
457 err = md_get_root(sbi->ll_md_exp, get_mount_fileset(sb),
460 CERROR("cannot mds_connect: rc = %d\n", err);
461 GOTO(out_lock_cn_cb, err);
463 if (!fid_is_sane(&sbi->ll_root_fid)) {
464 CERROR("%s: Invalid root fid "DFID" during mount\n",
465 sbi->ll_md_exp->exp_obd->obd_name,
466 PFID(&sbi->ll_root_fid));
467 GOTO(out_lock_cn_cb, err = -EINVAL);
469 CDEBUG(D_SUPER, "rootfid "DFID"\n", PFID(&sbi->ll_root_fid));
471 sb->s_op = &lustre_super_operations;
472 #if THREAD_SIZE >= 8192 /*b=17630*/
473 sb->s_export_op = &lustre_export_operations;
477 * XXX: move this to after cbd setup? */
478 valid = OBD_MD_FLGETATTR | OBD_MD_FLBLOCKS | OBD_MD_FLMODEASIZE;
479 if (sbi->ll_flags & LL_SBI_ACL)
480 valid |= OBD_MD_FLACL;
482 OBD_ALLOC_PTR(op_data);
484 GOTO(out_lock_cn_cb, err = -ENOMEM);
486 op_data->op_fid1 = sbi->ll_root_fid;
487 op_data->op_mode = 0;
488 op_data->op_valid = valid;
490 err = md_getattr(sbi->ll_md_exp, op_data, &request);
492 OBD_FREE_PTR(op_data);
494 CERROR("%s: md_getattr failed for root: rc = %d\n",
495 sbi->ll_md_exp->exp_obd->obd_name, err);
496 GOTO(out_lock_cn_cb, err);
499 err = md_get_lustre_md(sbi->ll_md_exp, request, sbi->ll_dt_exp,
500 sbi->ll_md_exp, &lmd);
502 CERROR("failed to understand root inode md: rc = %d\n", err);
503 ptlrpc_req_finished(request);
504 GOTO(out_lock_cn_cb, err);
507 LASSERT(fid_is_sane(&sbi->ll_root_fid));
508 root = ll_iget(sb, cl_fid_build_ino(&sbi->ll_root_fid,
509 sbi->ll_flags & LL_SBI_32BIT_API),
511 md_free_lustre_md(sbi->ll_md_exp, &lmd);
512 ptlrpc_req_finished(request);
515 #ifdef CONFIG_FS_POSIX_ACL
517 posix_acl_release(lmd.posix_acl);
518 lmd.posix_acl = NULL;
521 err = IS_ERR(root) ? PTR_ERR(root) : -EBADF;
523 CERROR("lustre_lite: bad iget4 for root\n");
527 checksum = sbi->ll_flags & LL_SBI_CHECKSUM;
528 err = obd_set_info_async(NULL, sbi->ll_dt_exp, sizeof(KEY_CHECKSUM),
529 KEY_CHECKSUM, sizeof(checksum), &checksum,
532 CERROR("%s: Set checksum failed: rc = %d\n",
533 sbi->ll_dt_exp->exp_obd->obd_name, err);
538 err = obd_set_info_async(NULL, sbi->ll_dt_exp, sizeof(KEY_CACHE_SET),
539 KEY_CACHE_SET, sizeof(*sbi->ll_cache),
540 sbi->ll_cache, NULL);
542 CERROR("%s: Set cache_set failed: rc = %d\n",
543 sbi->ll_dt_exp->exp_obd->obd_name, err);
547 sb->s_root = d_make_root(root);
548 if (sb->s_root == NULL) {
549 CERROR("%s: can't make root dentry\n",
550 ll_get_fsname(sb, NULL, 0));
551 GOTO(out_root, err = -ENOMEM);
553 #ifdef HAVE_DCACHE_LOCK
554 sb->s_root->d_op = &ll_d_ops;
557 sbi->ll_sdev_orig = sb->s_dev;
559 /* We set sb->s_dev equal on all lustre clients in order to support
560 * NFS export clustering. NFSD requires that the FSID be the same
562 /* s_dev is also used in lt_compare() to compare two fs, but that is
563 * only a node-local comparison. */
564 uuid = obd_get_uuid(sbi->ll_md_exp);
566 sb->s_dev = get_uuid2int(uuid->uuid, strlen(uuid->uuid));
572 if (proc_lustre_fs_root != NULL) {
573 err = lprocfs_register_mountpoint(proc_lustre_fs_root, sb,
576 CERROR("%s: could not register mount in lprocfs: "
577 "rc = %d\n", ll_get_fsname(sb, NULL, 0), err);
587 obd_fid_fini(sbi->ll_dt_exp->exp_obd);
589 obd_disconnect(sbi->ll_dt_exp);
590 sbi->ll_dt_exp = NULL;
592 obd_fid_fini(sbi->ll_md_exp->exp_obd);
594 obd_disconnect(sbi->ll_md_exp);
595 sbi->ll_md_exp = NULL;
604 int ll_get_max_mdsize(struct ll_sb_info *sbi, int *lmmsize)
608 size = sizeof(*lmmsize);
609 rc = obd_get_info(NULL, sbi->ll_dt_exp, sizeof(KEY_MAX_EASIZE),
610 KEY_MAX_EASIZE, &size, lmmsize);
612 CERROR("%s: cannot get max LOV EA size: rc = %d\n",
613 sbi->ll_dt_exp->exp_obd->obd_name, rc);
618 rc = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_MAX_EASIZE),
619 KEY_MAX_EASIZE, &size, lmmsize);
621 CERROR("Get max mdsize error rc %d\n", rc);
627 * Get the value of the default_easize parameter.
629 * \see client_obd::cl_default_mds_easize
631 * \param[in] sbi superblock info for this filesystem
632 * \param[out] lmmsize pointer to storage location for value
634 * \retval 0 on success
635 * \retval negative negated errno on failure
637 int ll_get_default_mdsize(struct ll_sb_info *sbi, int *lmmsize)
642 rc = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_DEFAULT_EASIZE),
643 KEY_DEFAULT_EASIZE, &size, lmmsize);
645 CERROR("Get default mdsize error rc %d\n", rc);
651 * Set the default_easize parameter to the given value.
653 * \see client_obd::cl_default_mds_easize
655 * \param[in] sbi superblock info for this filesystem
656 * \param[in] lmmsize the size to set
658 * \retval 0 on success
659 * \retval negative negated errno on failure
661 int ll_set_default_mdsize(struct ll_sb_info *sbi, int lmmsize)
665 if (lmmsize < sizeof(struct lov_mds_md) ||
666 lmmsize > OBD_MAX_DEFAULT_EA_SIZE)
669 rc = obd_set_info_async(NULL, sbi->ll_md_exp,
670 sizeof(KEY_DEFAULT_EASIZE), KEY_DEFAULT_EASIZE,
671 sizeof(int), &lmmsize, NULL);
676 static void client_common_put_super(struct super_block *sb)
678 struct ll_sb_info *sbi = ll_s2sbi(sb);
683 obd_fid_fini(sbi->ll_dt_exp->exp_obd);
684 obd_disconnect(sbi->ll_dt_exp);
685 sbi->ll_dt_exp = NULL;
687 lprocfs_unregister_mountpoint(sbi);
689 obd_fid_fini(sbi->ll_md_exp->exp_obd);
690 obd_disconnect(sbi->ll_md_exp);
691 sbi->ll_md_exp = NULL;
696 void ll_kill_super(struct super_block *sb)
698 struct ll_sb_info *sbi;
702 if (!(sb->s_flags & MS_ACTIVE))
706 /* we need restore s_dev from changed for clustred NFS before put_super
707 * because new kernels have cached s_dev and change sb->s_dev in
708 * put_super not affected real removing devices */
710 sb->s_dev = sbi->ll_sdev_orig;
711 sbi->ll_umounting = 1;
713 /* wait running statahead threads to quit */
714 while (atomic_read(&sbi->ll_sa_running) > 0) {
715 set_current_state(TASK_UNINTERRUPTIBLE);
716 schedule_timeout(msecs_to_jiffies(MSEC_PER_SEC >> 3));
723 static inline int ll_set_opt(const char *opt, char *data, int fl)
725 if (strncmp(opt, data, strlen(opt)) != 0)
731 /* non-client-specific mount options are parsed in lmd_parse */
732 static int ll_options(char *options, int *flags)
735 char *s1 = options, *s2;
741 CDEBUG(D_CONFIG, "Parsing opts %s\n", options);
744 CDEBUG(D_SUPER, "next opt=%s\n", s1);
745 tmp = ll_set_opt("nolock", s1, LL_SBI_NOLCK);
750 tmp = ll_set_opt("flock", s1, LL_SBI_FLOCK);
755 tmp = ll_set_opt("localflock", s1, LL_SBI_LOCALFLOCK);
760 tmp = ll_set_opt("noflock", s1, LL_SBI_FLOCK|LL_SBI_LOCALFLOCK);
765 tmp = ll_set_opt("user_xattr", s1, LL_SBI_USER_XATTR);
770 tmp = ll_set_opt("nouser_xattr", s1, LL_SBI_USER_XATTR);
775 tmp = ll_set_opt("context", s1, 1);
778 tmp = ll_set_opt("fscontext", s1, 1);
781 tmp = ll_set_opt("defcontext", s1, 1);
784 tmp = ll_set_opt("rootcontext", s1, 1);
787 tmp = ll_set_opt("user_fid2path", s1, LL_SBI_USER_FID2PATH);
792 tmp = ll_set_opt("nouser_fid2path", s1, LL_SBI_USER_FID2PATH);
798 tmp = ll_set_opt("checksum", s1, LL_SBI_CHECKSUM);
803 tmp = ll_set_opt("nochecksum", s1, LL_SBI_CHECKSUM);
808 tmp = ll_set_opt("lruresize", s1, LL_SBI_LRU_RESIZE);
813 tmp = ll_set_opt("nolruresize", s1, LL_SBI_LRU_RESIZE);
818 tmp = ll_set_opt("lazystatfs", s1, LL_SBI_LAZYSTATFS);
823 tmp = ll_set_opt("nolazystatfs", s1, LL_SBI_LAZYSTATFS);
828 tmp = ll_set_opt("32bitapi", s1, LL_SBI_32BIT_API);
833 tmp = ll_set_opt("verbose", s1, LL_SBI_VERBOSE);
838 tmp = ll_set_opt("noverbose", s1, LL_SBI_VERBOSE);
843 tmp = ll_set_opt("always_ping", s1, LL_SBI_ALWAYS_PING);
848 LCONSOLE_ERROR_MSG(0x152, "Unknown option '%s', won't mount.\n",
854 s2 = strchr(s1, ',');
862 void ll_lli_init(struct ll_inode_info *lli)
864 lli->lli_inode_magic = LLI_INODE_MAGIC;
866 spin_lock_init(&lli->lli_lock);
867 lli->lli_posix_acl = NULL;
868 /* Do not set lli_fid, it has been initialized already. */
869 fid_zero(&lli->lli_pfid);
870 lli->lli_mds_read_och = NULL;
871 lli->lli_mds_write_och = NULL;
872 lli->lli_mds_exec_och = NULL;
873 lli->lli_open_fd_read_count = 0;
874 lli->lli_open_fd_write_count = 0;
875 lli->lli_open_fd_exec_count = 0;
876 mutex_init(&lli->lli_och_mutex);
877 spin_lock_init(&lli->lli_agl_lock);
878 spin_lock_init(&lli->lli_layout_lock);
879 ll_layout_version_set(lli, CL_LAYOUT_GEN_NONE);
880 lli->lli_clob = NULL;
882 init_rwsem(&lli->lli_xattrs_list_rwsem);
883 mutex_init(&lli->lli_xattrs_enq_lock);
885 LASSERT(lli->lli_vfs_inode.i_mode != 0);
886 if (S_ISDIR(lli->lli_vfs_inode.i_mode)) {
887 mutex_init(&lli->lli_readdir_mutex);
888 lli->lli_opendir_key = NULL;
890 spin_lock_init(&lli->lli_sa_lock);
891 lli->lli_opendir_pid = 0;
892 lli->lli_sa_enabled = 0;
893 lli->lli_def_stripe_offset = -1;
895 mutex_init(&lli->lli_size_mutex);
896 lli->lli_symlink_name = NULL;
897 init_rwsem(&lli->lli_trunc_sem);
898 range_lock_tree_init(&lli->lli_write_tree);
899 init_rwsem(&lli->lli_glimpse_sem);
900 lli->lli_glimpse_time = 0;
901 INIT_LIST_HEAD(&lli->lli_agl_list);
902 lli->lli_agl_index = 0;
903 lli->lli_async_rc = 0;
905 mutex_init(&lli->lli_layout_mutex);
906 memset(lli->lli_jobid, 0, LUSTRE_JOBID_SIZE);
909 static inline int ll_bdi_register(struct backing_dev_info *bdi)
911 static atomic_t ll_bdi_num = ATOMIC_INIT(0);
913 bdi->name = "lustre";
914 return bdi_register(bdi, NULL, "lustre-%d",
915 atomic_inc_return(&ll_bdi_num));
918 int ll_fill_super(struct super_block *sb, struct vfsmount *mnt)
920 struct lustre_profile *lprof = NULL;
921 struct lustre_sb_info *lsi = s2lsi(sb);
922 struct ll_sb_info *sbi;
923 char *dt = NULL, *md = NULL;
924 char *profilenm = get_profile_name(sb);
925 struct config_llog_instance *cfg;
926 /* %p for void* in printf needs 16+2 characters: 0xffffffffffffffff */
927 const int instlen = sizeof(cfg->cfg_instance) * 2 + 2;
931 CDEBUG(D_VFSTRACE, "VFS Op: sb %p\n", sb);
937 try_module_get(THIS_MODULE);
939 /* client additional sb info */
940 lsi->lsi_llsbi = sbi = ll_init_sbi();
942 module_put(THIS_MODULE);
947 err = ll_options(lsi->lsi_lmd->lmd_opts, &sbi->ll_flags);
951 err = bdi_init(&lsi->lsi_bdi);
954 lsi->lsi_flags |= LSI_BDI_INITIALIZED;
955 #ifdef HAVE_BDI_CAP_MAP_COPY
956 lsi->lsi_bdi.capabilities = BDI_CAP_MAP_COPY;
958 lsi->lsi_bdi.capabilities = 0;
960 err = ll_bdi_register(&lsi->lsi_bdi);
964 sb->s_bdi = &lsi->lsi_bdi;
965 #ifndef HAVE_DCACHE_LOCK
966 /* kernel >= 2.6.38 store dentry operations in sb->s_d_op. */
967 sb->s_d_op = &ll_d_ops;
970 /* Generate a string unique to this super, in case some joker tries
971 to mount the same fs at two mount points.
972 Use the address of the super itself.*/
973 cfg->cfg_instance = sb;
974 cfg->cfg_uuid = lsi->lsi_llsbi->ll_sb_uuid;
975 cfg->cfg_callback = class_config_llog_handler;
976 cfg->cfg_sub_clds = CONFIG_SUB_CLIENT;
977 /* set up client obds */
978 err = lustre_process_log(sb, profilenm, cfg);
982 /* Profile set with LCFG_MOUNTOPT so we can find our mdc and osc obds */
983 lprof = class_get_profile(profilenm);
985 LCONSOLE_ERROR_MSG(0x156, "The client profile '%s' could not be"
986 " read from the MGS. Does that filesystem "
987 "exist?\n", profilenm);
988 GOTO(out_free, err = -EINVAL);
990 CDEBUG(D_CONFIG, "Found profile %s: mdc=%s osc=%s\n", profilenm,
991 lprof->lp_md, lprof->lp_dt);
993 OBD_ALLOC(dt, strlen(lprof->lp_dt) + instlen + 2);
995 GOTO(out_free, err = -ENOMEM);
996 sprintf(dt, "%s-%p", lprof->lp_dt, cfg->cfg_instance);
998 OBD_ALLOC(md, strlen(lprof->lp_md) + instlen + 2);
1000 GOTO(out_free, err = -ENOMEM);
1001 sprintf(md, "%s-%p", lprof->lp_md, cfg->cfg_instance);
1003 /* connections, registrations, sb setup */
1004 err = client_common_fill_super(sb, md, dt, mnt);
1006 GOTO(out_free, err);
1008 sbi->ll_client_common_fill_super_succeeded = 1;
1012 OBD_FREE(md, strlen(lprof->lp_md) + instlen + 2);
1014 OBD_FREE(dt, strlen(lprof->lp_dt) + instlen + 2);
1016 class_put_profile(lprof);
1019 else if (sbi->ll_flags & LL_SBI_VERBOSE)
1020 LCONSOLE_WARN("Mounted %s\n", profilenm);
1024 } /* ll_fill_super */
1026 void ll_put_super(struct super_block *sb)
1028 struct config_llog_instance cfg, params_cfg;
1029 struct obd_device *obd;
1030 struct lustre_sb_info *lsi = s2lsi(sb);
1031 struct ll_sb_info *sbi = ll_s2sbi(sb);
1032 char *profilenm = get_profile_name(sb);
1034 int next, force = 1, rc = 0;
1037 CDEBUG(D_VFSTRACE, "VFS Op: sb %p - %s\n", sb, profilenm);
1039 cfg.cfg_instance = sb;
1040 lustre_end_log(sb, profilenm, &cfg);
1042 params_cfg.cfg_instance = sb;
1043 lustre_end_log(sb, PARAMS_FILENAME, ¶ms_cfg);
1045 if (sbi->ll_md_exp) {
1046 obd = class_exp2obd(sbi->ll_md_exp);
1048 force = obd->obd_force;
1051 /* Wait for unstable pages to be committed to stable storage */
1053 struct l_wait_info lwi = LWI_INTR(LWI_ON_SIGNAL_NOOP, NULL);
1054 rc = l_wait_event(sbi->ll_cache->ccc_unstable_waitq,
1055 atomic_long_read(&sbi->ll_cache->ccc_unstable_nr) == 0,
1059 ccc_count = atomic_long_read(&sbi->ll_cache->ccc_unstable_nr);
1060 if (force == 0 && rc != -EINTR)
1061 LASSERTF(ccc_count == 0, "count: %li\n", ccc_count);
1064 /* We need to set force before the lov_disconnect in
1065 lustre_common_put_super, since l_d cleans up osc's as well. */
1068 while ((obd = class_devices_in_group(&sbi->ll_sb_uuid,
1070 obd->obd_force = force;
1074 if (sbi->ll_client_common_fill_super_succeeded) {
1075 /* Only if client_common_fill_super succeeded */
1076 client_common_put_super(sb);
1080 while ((obd = class_devices_in_group(&sbi->ll_sb_uuid, &next)) !=NULL) {
1081 class_manual_cleanup(obd);
1084 if (sbi->ll_flags & LL_SBI_VERBOSE)
1085 LCONSOLE_WARN("Unmounted %s\n", profilenm ? profilenm : "");
1088 class_del_profile(profilenm);
1090 if (lsi->lsi_flags & LSI_BDI_INITIALIZED) {
1091 bdi_destroy(&lsi->lsi_bdi);
1092 lsi->lsi_flags &= ~LSI_BDI_INITIALIZED;
1096 lsi->lsi_llsbi = NULL;
1098 lustre_common_put_super(sb);
1100 cl_env_cache_purge(~0);
1102 module_put(THIS_MODULE);
1105 } /* client_put_super */
1107 struct inode *ll_inode_from_resource_lock(struct ldlm_lock *lock)
1109 struct inode *inode = NULL;
1111 /* NOTE: we depend on atomic igrab() -bzzz */
1112 lock_res_and_lock(lock);
1113 if (lock->l_resource->lr_lvb_inode) {
1114 struct ll_inode_info * lli;
1115 lli = ll_i2info(lock->l_resource->lr_lvb_inode);
1116 if (lli->lli_inode_magic == LLI_INODE_MAGIC) {
1117 inode = igrab(lock->l_resource->lr_lvb_inode);
1119 inode = lock->l_resource->lr_lvb_inode;
1120 LDLM_DEBUG_LIMIT(inode->i_state & I_FREEING ? D_INFO :
1121 D_WARNING, lock, "lr_lvb_inode %p is "
1122 "bogus: magic %08x",
1123 lock->l_resource->lr_lvb_inode,
1124 lli->lli_inode_magic);
1128 unlock_res_and_lock(lock);
1132 void ll_dir_clear_lsm_md(struct inode *inode)
1134 struct ll_inode_info *lli = ll_i2info(inode);
1136 LASSERT(S_ISDIR(inode->i_mode));
1138 if (lli->lli_lsm_md != NULL) {
1139 lmv_free_memmd(lli->lli_lsm_md);
1140 lli->lli_lsm_md = NULL;
1144 static struct inode *ll_iget_anon_dir(struct super_block *sb,
1145 const struct lu_fid *fid,
1146 struct lustre_md *md)
1148 struct ll_sb_info *sbi = ll_s2sbi(sb);
1149 struct mdt_body *body = md->body;
1150 struct inode *inode;
1154 ino = cl_fid_build_ino(fid, sbi->ll_flags & LL_SBI_32BIT_API);
1155 inode = iget_locked(sb, ino);
1156 if (inode == NULL) {
1157 CERROR("%s: failed get simple inode "DFID": rc = -ENOENT\n",
1158 ll_get_fsname(sb, NULL, 0), PFID(fid));
1159 RETURN(ERR_PTR(-ENOENT));
1162 if (inode->i_state & I_NEW) {
1163 struct ll_inode_info *lli = ll_i2info(inode);
1164 struct lmv_stripe_md *lsm = md->lmv;
1166 inode->i_mode = (inode->i_mode & ~S_IFMT) |
1167 (body->mbo_mode & S_IFMT);
1168 LASSERTF(S_ISDIR(inode->i_mode), "Not slave inode "DFID"\n",
1171 LTIME_S(inode->i_mtime) = 0;
1172 LTIME_S(inode->i_atime) = 0;
1173 LTIME_S(inode->i_ctime) = 0;
1176 #ifdef HAVE_BACKING_DEV_INFO
1177 /* initializing backing dev info. */
1178 inode->i_mapping->backing_dev_info =
1179 &s2lsi(inode->i_sb)->lsi_bdi;
1181 inode->i_op = &ll_dir_inode_operations;
1182 inode->i_fop = &ll_dir_operations;
1183 lli->lli_fid = *fid;
1186 LASSERT(lsm != NULL);
1187 /* master object FID */
1188 lli->lli_pfid = body->mbo_fid1;
1189 CDEBUG(D_INODE, "lli %p slave "DFID" master "DFID"\n",
1190 lli, PFID(fid), PFID(&lli->lli_pfid));
1191 unlock_new_inode(inode);
1197 static int ll_init_lsm_md(struct inode *inode, struct lustre_md *md)
1200 struct lmv_stripe_md *lsm = md->lmv;
1203 LASSERT(lsm != NULL);
1204 /* XXX sigh, this lsm_root initialization should be in
1205 * LMV layer, but it needs ll_iget right now, so we
1206 * put this here right now. */
1207 for (i = 0; i < lsm->lsm_md_stripe_count; i++) {
1208 fid = &lsm->lsm_md_oinfo[i].lmo_fid;
1209 LASSERT(lsm->lsm_md_oinfo[i].lmo_root == NULL);
1210 /* Unfortunately ll_iget will call ll_update_inode,
1211 * where the initialization of slave inode is slightly
1212 * different, so it reset lsm_md to NULL to avoid
1213 * initializing lsm for slave inode. */
1214 /* For migrating inode, master stripe and master object will
1215 * be same, so we only need assign this inode */
1216 if (lsm->lsm_md_hash_type & LMV_HASH_FLAG_MIGRATION && i == 0)
1217 lsm->lsm_md_oinfo[i].lmo_root = inode;
1219 lsm->lsm_md_oinfo[i].lmo_root =
1220 ll_iget_anon_dir(inode->i_sb, fid, md);
1222 if (IS_ERR(lsm->lsm_md_oinfo[i].lmo_root)) {
1223 int rc = PTR_ERR(lsm->lsm_md_oinfo[i].lmo_root);
1225 lsm->lsm_md_oinfo[i].lmo_root = NULL;
1233 static inline int lli_lsm_md_eq(const struct lmv_stripe_md *lsm_md1,
1234 const struct lmv_stripe_md *lsm_md2)
1236 return lsm_md1->lsm_md_magic == lsm_md2->lsm_md_magic &&
1237 lsm_md1->lsm_md_stripe_count == lsm_md2->lsm_md_stripe_count &&
1238 lsm_md1->lsm_md_master_mdt_index ==
1239 lsm_md2->lsm_md_master_mdt_index &&
1240 lsm_md1->lsm_md_hash_type == lsm_md2->lsm_md_hash_type &&
1241 lsm_md1->lsm_md_layout_version ==
1242 lsm_md2->lsm_md_layout_version &&
1243 strcmp(lsm_md1->lsm_md_pool_name,
1244 lsm_md2->lsm_md_pool_name) == 0;
1247 static int ll_update_lsm_md(struct inode *inode, struct lustre_md *md)
1249 struct ll_inode_info *lli = ll_i2info(inode);
1250 struct lmv_stripe_md *lsm = md->lmv;
1254 LASSERT(S_ISDIR(inode->i_mode));
1255 CDEBUG(D_INODE, "update lsm %p of "DFID"\n", lli->lli_lsm_md,
1256 PFID(ll_inode2fid(inode)));
1258 /* no striped information from request. */
1260 if (lli->lli_lsm_md == NULL) {
1262 } else if (lli->lli_lsm_md->lsm_md_hash_type &
1263 LMV_HASH_FLAG_MIGRATION) {
1264 /* migration is done, the temporay MIGRATE layout has
1266 CDEBUG(D_INODE, DFID" finish migration.\n",
1267 PFID(ll_inode2fid(inode)));
1268 lmv_free_memmd(lli->lli_lsm_md);
1269 lli->lli_lsm_md = NULL;
1272 /* The lustre_md from req does not include stripeEA,
1273 * see ll_md_setattr */
1278 /* set the directory layout */
1279 if (lli->lli_lsm_md == NULL) {
1280 struct cl_attr *attr;
1282 rc = ll_init_lsm_md(inode, md);
1286 /* set md->lmv to NULL, so the following free lustre_md
1287 * will not free this lsm */
1289 lli->lli_lsm_md = lsm;
1291 OBD_ALLOC_PTR(attr);
1295 /* validate the lsm */
1296 rc = md_merge_attr(ll_i2mdexp(inode), lsm, attr,
1297 ll_md_blocking_ast);
1303 if (md->body->mbo_valid & OBD_MD_FLNLINK)
1304 md->body->mbo_nlink = attr->cat_nlink;
1305 if (md->body->mbo_valid & OBD_MD_FLSIZE)
1306 md->body->mbo_size = attr->cat_size;
1307 if (md->body->mbo_valid & OBD_MD_FLATIME)
1308 md->body->mbo_atime = attr->cat_atime;
1309 if (md->body->mbo_valid & OBD_MD_FLCTIME)
1310 md->body->mbo_ctime = attr->cat_ctime;
1311 if (md->body->mbo_valid & OBD_MD_FLMTIME)
1312 md->body->mbo_mtime = attr->cat_mtime;
1316 CDEBUG(D_INODE, "Set lsm %p magic %x to "DFID"\n", lsm,
1317 lsm->lsm_md_magic, PFID(ll_inode2fid(inode)));
1321 /* Compare the old and new stripe information */
1322 if (!lsm_md_eq(lli->lli_lsm_md, lsm)) {
1323 struct lmv_stripe_md *old_lsm = lli->lli_lsm_md;
1326 CERROR("%s: inode "DFID"(%p)'s lmv layout mismatch (%p)/(%p)"
1327 "magic:0x%x/0x%x stripe count: %d/%d master_mdt: %d/%d"
1328 "hash_type:0x%x/0x%x layout: 0x%x/0x%x pool:%s/%s\n",
1329 ll_get_fsname(inode->i_sb, NULL, 0), PFID(&lli->lli_fid),
1330 inode, lsm, old_lsm,
1331 lsm->lsm_md_magic, old_lsm->lsm_md_magic,
1332 lsm->lsm_md_stripe_count,
1333 old_lsm->lsm_md_stripe_count,
1334 lsm->lsm_md_master_mdt_index,
1335 old_lsm->lsm_md_master_mdt_index,
1336 lsm->lsm_md_hash_type, old_lsm->lsm_md_hash_type,
1337 lsm->lsm_md_layout_version,
1338 old_lsm->lsm_md_layout_version,
1339 lsm->lsm_md_pool_name,
1340 old_lsm->lsm_md_pool_name);
1342 for (idx = 0; idx < old_lsm->lsm_md_stripe_count; idx++) {
1343 CERROR("%s: sub FIDs in old lsm idx %d, old: "DFID"\n",
1344 ll_get_fsname(inode->i_sb, NULL, 0), idx,
1345 PFID(&old_lsm->lsm_md_oinfo[idx].lmo_fid));
1348 for (idx = 0; idx < lsm->lsm_md_stripe_count; idx++) {
1349 CERROR("%s: sub FIDs in new lsm idx %d, new: "DFID"\n",
1350 ll_get_fsname(inode->i_sb, NULL, 0), idx,
1351 PFID(&lsm->lsm_md_oinfo[idx].lmo_fid));
1360 void ll_clear_inode(struct inode *inode)
1362 struct ll_inode_info *lli = ll_i2info(inode);
1363 struct ll_sb_info *sbi = ll_i2sbi(inode);
1366 CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p)\n",
1367 PFID(ll_inode2fid(inode)), inode);
1369 if (S_ISDIR(inode->i_mode)) {
1370 /* these should have been cleared in ll_file_release */
1371 LASSERT(lli->lli_opendir_key == NULL);
1372 LASSERT(lli->lli_sai == NULL);
1373 LASSERT(lli->lli_opendir_pid == 0);
1376 md_null_inode(sbi->ll_md_exp, ll_inode2fid(inode));
1378 LASSERT(!lli->lli_open_fd_write_count);
1379 LASSERT(!lli->lli_open_fd_read_count);
1380 LASSERT(!lli->lli_open_fd_exec_count);
1382 if (lli->lli_mds_write_och)
1383 ll_md_real_close(inode, FMODE_WRITE);
1384 if (lli->lli_mds_exec_och)
1385 ll_md_real_close(inode, FMODE_EXEC);
1386 if (lli->lli_mds_read_och)
1387 ll_md_real_close(inode, FMODE_READ);
1389 if (S_ISLNK(inode->i_mode) && lli->lli_symlink_name) {
1390 OBD_FREE(lli->lli_symlink_name,
1391 strlen(lli->lli_symlink_name) + 1);
1392 lli->lli_symlink_name = NULL;
1395 ll_xattr_cache_destroy(inode);
1397 #ifdef CONFIG_FS_POSIX_ACL
1398 if (lli->lli_posix_acl) {
1399 LASSERT(atomic_read(&lli->lli_posix_acl->a_refcount) == 1);
1400 posix_acl_release(lli->lli_posix_acl);
1401 lli->lli_posix_acl = NULL;
1404 lli->lli_inode_magic = LLI_INODE_DEAD;
1406 if (S_ISDIR(inode->i_mode))
1407 ll_dir_clear_lsm_md(inode);
1408 else if (S_ISREG(inode->i_mode) && !is_bad_inode(inode))
1409 LASSERT(list_empty(&lli->lli_agl_list));
1412 * XXX This has to be done before lsm is freed below, because
1413 * cl_object still uses inode lsm.
1415 cl_inode_fini(inode);
1420 static int ll_md_setattr(struct dentry *dentry, struct md_op_data *op_data)
1422 struct lustre_md md;
1423 struct inode *inode = dentry->d_inode;
1424 struct ll_sb_info *sbi = ll_i2sbi(inode);
1425 struct ptlrpc_request *request = NULL;
1429 op_data = ll_prep_md_op_data(op_data, inode, NULL, NULL, 0, 0,
1430 LUSTRE_OPC_ANY, NULL);
1431 if (IS_ERR(op_data))
1432 RETURN(PTR_ERR(op_data));
1434 rc = md_setattr(sbi->ll_md_exp, op_data, NULL, 0, &request);
1436 ptlrpc_req_finished(request);
1437 if (rc == -ENOENT) {
1439 /* Unlinked special device node? Or just a race?
1440 * Pretend we done everything. */
1441 if (!S_ISREG(inode->i_mode) &&
1442 !S_ISDIR(inode->i_mode)) {
1443 ia_valid = op_data->op_attr.ia_valid;
1444 op_data->op_attr.ia_valid &= ~TIMES_SET_FLAGS;
1445 rc = simple_setattr(dentry, &op_data->op_attr);
1446 op_data->op_attr.ia_valid = ia_valid;
1448 } else if (rc != -EPERM && rc != -EACCES && rc != -ETXTBSY) {
1449 CERROR("md_setattr fails: rc = %d\n", rc);
1454 rc = md_get_lustre_md(sbi->ll_md_exp, request, sbi->ll_dt_exp,
1455 sbi->ll_md_exp, &md);
1457 ptlrpc_req_finished(request);
1461 ia_valid = op_data->op_attr.ia_valid;
1462 /* inode size will be in ll_setattr_ost, can't do it now since dirty
1463 * cache is not cleared yet. */
1464 op_data->op_attr.ia_valid &= ~(TIMES_SET_FLAGS | ATTR_SIZE);
1465 if (S_ISREG(inode->i_mode))
1467 rc = simple_setattr(dentry, &op_data->op_attr);
1468 if (S_ISREG(inode->i_mode))
1469 inode_unlock(inode);
1470 op_data->op_attr.ia_valid = ia_valid;
1472 rc = ll_update_inode(inode, &md);
1473 ptlrpc_req_finished(request);
1478 /* If this inode has objects allocated to it (lsm != NULL), then the OST
1479 * object(s) determine the file size and mtime. Otherwise, the MDS will
1480 * keep these values until such a time that objects are allocated for it.
1481 * We do the MDS operations first, as it is checking permissions for us.
1482 * We don't to the MDS RPC if there is nothing that we want to store there,
1483 * otherwise there is no harm in updating mtime/atime on the MDS if we are
1484 * going to do an RPC anyways.
1486 * If we are doing a truncate, we will send the mtime and ctime updates
1487 * to the OST with the punch RPC, otherwise we do an explicit setattr RPC.
1488 * I don't believe it is possible to get e.g. ATTR_MTIME_SET and ATTR_SIZE
1491 * In case of HSMimport, we only set attr on MDS.
1493 int ll_setattr_raw(struct dentry *dentry, struct iattr *attr, bool hsm_import)
1495 struct inode *inode = dentry->d_inode;
1496 struct ll_inode_info *lli = ll_i2info(inode);
1497 struct md_op_data *op_data = NULL;
1501 CDEBUG(D_VFSTRACE, "%s: setattr inode "DFID"(%p) from %llu to %llu, "
1502 "valid %x, hsm_import %d\n",
1503 ll_get_fsname(inode->i_sb, NULL, 0), PFID(&lli->lli_fid),
1504 inode, i_size_read(inode), attr->ia_size, attr->ia_valid,
1507 if (attr->ia_valid & ATTR_SIZE) {
1508 /* Check new size against VFS/VM file size limit and rlimit */
1509 rc = inode_newsize_ok(inode, attr->ia_size);
1513 /* The maximum Lustre file size is variable, based on the
1514 * OST maximum object size and number of stripes. This
1515 * needs another check in addition to the VFS check above. */
1516 if (attr->ia_size > ll_file_maxbytes(inode)) {
1517 CDEBUG(D_INODE,"file "DFID" too large %llu > %llu\n",
1518 PFID(&lli->lli_fid), attr->ia_size,
1519 ll_file_maxbytes(inode));
1523 attr->ia_valid |= ATTR_MTIME | ATTR_CTIME;
1526 /* POSIX: check before ATTR_*TIME_SET set (from inode_change_ok) */
1527 if (attr->ia_valid & TIMES_SET_FLAGS) {
1528 if ((!uid_eq(current_fsuid(), inode->i_uid)) &&
1529 !cfs_capable(CFS_CAP_FOWNER))
1533 /* We mark all of the fields "set" so MDS/OST does not re-set them */
1534 if (!(attr->ia_valid & ATTR_CTIME_SET) &&
1535 (attr->ia_valid & ATTR_CTIME)) {
1536 attr->ia_ctime = CURRENT_TIME;
1537 attr->ia_valid |= ATTR_CTIME_SET;
1539 if (!(attr->ia_valid & ATTR_ATIME_SET) &&
1540 (attr->ia_valid & ATTR_ATIME)) {
1541 attr->ia_atime = CURRENT_TIME;
1542 attr->ia_valid |= ATTR_ATIME_SET;
1544 if (!(attr->ia_valid & ATTR_MTIME_SET) &&
1545 (attr->ia_valid & ATTR_MTIME)) {
1546 attr->ia_mtime = CURRENT_TIME;
1547 attr->ia_valid |= ATTR_MTIME_SET;
1550 if (attr->ia_valid & (ATTR_MTIME | ATTR_CTIME))
1551 CDEBUG(D_INODE, "setting mtime %lu, ctime %lu, now = %llu\n",
1552 LTIME_S(attr->ia_mtime), LTIME_S(attr->ia_ctime),
1553 (s64)ktime_get_real_seconds());
1555 if (S_ISREG(inode->i_mode)) {
1556 if (attr->ia_valid & ATTR_SIZE)
1557 inode_dio_write_done(inode);
1558 inode_unlock(inode);
1561 /* We always do an MDS RPC, even if we're only changing the size;
1562 * only the MDS knows whether truncate() should fail with -ETXTBUSY */
1564 OBD_ALLOC_PTR(op_data);
1565 if (op_data == NULL)
1566 GOTO(out, rc = -ENOMEM);
1568 if (!hsm_import && attr->ia_valid & ATTR_SIZE) {
1569 /* If we are changing file size, file content is
1570 * modified, flag it. */
1571 attr->ia_valid |= MDS_OPEN_OWNEROVERRIDE;
1572 op_data->op_bias |= MDS_DATA_MODIFIED;
1573 ll_file_clear_flag(lli, LLIF_DATA_MODIFIED);
1576 op_data->op_attr = *attr;
1578 rc = ll_md_setattr(dentry, op_data);
1582 if (!S_ISREG(inode->i_mode) || hsm_import)
1585 if (attr->ia_valid & (ATTR_SIZE |
1586 ATTR_ATIME | ATTR_ATIME_SET |
1587 ATTR_MTIME | ATTR_MTIME_SET |
1588 ATTR_CTIME | ATTR_CTIME_SET)) {
1589 /* For truncate and utimes sending attributes to OSTs, setting
1590 * mtime/atime to the past will be performed under PW [0:EOF]
1591 * extent lock (new_size:EOF for truncate). It may seem
1592 * excessive to send mtime/atime updates to OSTs when not
1593 * setting times to past, but it is necessary due to possible
1594 * time de-synchronization between MDT inode and OST objects */
1595 rc = cl_setattr_ost(lli->lli_clob, attr, 0);
1598 /* If the file was restored, it needs to set dirty flag.
1600 * We've already sent MDS_DATA_MODIFIED flag in
1601 * ll_md_setattr() for truncate. However, the MDT refuses to
1602 * set the HS_DIRTY flag on released files, so we have to set
1603 * it again if the file has been restored. Please check how
1604 * LLIF_DATA_MODIFIED is set in vvp_io_setattr_fini().
1606 * Please notice that if the file is not released, the previous
1607 * MDS_DATA_MODIFIED has taken effect and usually
1608 * LLIF_DATA_MODIFIED is not set(see vvp_io_setattr_fini()).
1609 * This way we can save an RPC for common open + trunc
1611 if (ll_file_test_and_clear_flag(lli, LLIF_DATA_MODIFIED)) {
1612 struct hsm_state_set hss = {
1613 .hss_valid = HSS_SETMASK,
1614 .hss_setmask = HS_DIRTY,
1618 rc2 = ll_hsm_state_set(inode, &hss);
1619 /* truncate and write can happen at the same time, so that
1620 * the file can be set modified even though the file is not
1621 * restored from released state, and ll_hsm_state_set() is
1622 * not applicable for the file, and rc2 < 0 is normal in this
1625 CDEBUG(D_INFO, DFID "HSM set dirty failed: rc2 = %d\n",
1626 PFID(ll_inode2fid(inode)), rc2);
1631 if (op_data != NULL)
1632 ll_finish_md_op_data(op_data);
1634 if (S_ISREG(inode->i_mode)) {
1636 if ((attr->ia_valid & ATTR_SIZE) && !hsm_import)
1637 inode_dio_wait(inode);
1640 ll_stats_ops_tally(ll_i2sbi(inode), (attr->ia_valid & ATTR_SIZE) ?
1641 LPROC_LL_TRUNC : LPROC_LL_SETATTR, 1);
1646 int ll_setattr(struct dentry *de, struct iattr *attr)
1648 int mode = de->d_inode->i_mode;
1650 if ((attr->ia_valid & (ATTR_CTIME|ATTR_SIZE|ATTR_MODE)) ==
1651 (ATTR_CTIME|ATTR_SIZE|ATTR_MODE))
1652 attr->ia_valid |= MDS_OPEN_OWNEROVERRIDE;
1654 if (((attr->ia_valid & (ATTR_MODE|ATTR_FORCE|ATTR_SIZE)) ==
1655 (ATTR_SIZE|ATTR_MODE)) &&
1656 (((mode & S_ISUID) && !(attr->ia_mode & S_ISUID)) ||
1657 (((mode & (S_ISGID|S_IXGRP)) == (S_ISGID|S_IXGRP)) &&
1658 !(attr->ia_mode & S_ISGID))))
1659 attr->ia_valid |= ATTR_FORCE;
1661 if ((attr->ia_valid & ATTR_MODE) &&
1663 !(attr->ia_mode & S_ISUID) &&
1664 !(attr->ia_valid & ATTR_KILL_SUID))
1665 attr->ia_valid |= ATTR_KILL_SUID;
1667 if ((attr->ia_valid & ATTR_MODE) &&
1668 ((mode & (S_ISGID|S_IXGRP)) == (S_ISGID|S_IXGRP)) &&
1669 !(attr->ia_mode & S_ISGID) &&
1670 !(attr->ia_valid & ATTR_KILL_SGID))
1671 attr->ia_valid |= ATTR_KILL_SGID;
1673 /* avoid polluted from ATTR_TIMES_SET,
1674 * projid is not expected to be set here */
1675 attr->ia_valid &= ~MDS_ATTR_PROJID;
1677 return ll_setattr_raw(de, attr, false);
1680 int ll_statfs_internal(struct super_block *sb, struct obd_statfs *osfs,
1681 __u64 max_age, __u32 flags)
1683 struct ll_sb_info *sbi = ll_s2sbi(sb);
1684 struct obd_statfs obd_osfs;
1688 rc = obd_statfs(NULL, sbi->ll_md_exp, osfs, max_age, flags);
1690 CERROR("md_statfs fails: rc = %d\n", rc);
1694 osfs->os_type = sb->s_magic;
1696 CDEBUG(D_SUPER, "MDC blocks %llu/%llu objects %llu/%llu\n",
1697 osfs->os_bavail, osfs->os_blocks, osfs->os_ffree,osfs->os_files);
1699 if (sbi->ll_flags & LL_SBI_LAZYSTATFS)
1700 flags |= OBD_STATFS_NODELAY;
1702 rc = obd_statfs_rqset(sbi->ll_dt_exp, &obd_osfs, max_age, flags);
1704 CERROR("obd_statfs fails: rc = %d\n", rc);
1708 CDEBUG(D_SUPER, "OSC blocks %llu/%llu objects %llu/%llu\n",
1709 obd_osfs.os_bavail, obd_osfs.os_blocks, obd_osfs.os_ffree,
1712 osfs->os_bsize = obd_osfs.os_bsize;
1713 osfs->os_blocks = obd_osfs.os_blocks;
1714 osfs->os_bfree = obd_osfs.os_bfree;
1715 osfs->os_bavail = obd_osfs.os_bavail;
1717 /* If we don't have as many objects free on the OST as inodes
1718 * on the MDS, we reduce the total number of inodes to
1719 * compensate, so that the "inodes in use" number is correct.
1721 if (obd_osfs.os_ffree < osfs->os_ffree) {
1722 osfs->os_files = (osfs->os_files - osfs->os_ffree) +
1724 osfs->os_ffree = obd_osfs.os_ffree;
1729 int ll_statfs(struct dentry *de, struct kstatfs *sfs)
1731 struct super_block *sb = de->d_sb;
1732 struct obd_statfs osfs;
1733 __u64 fsid = huge_encode_dev(sb->s_dev);
1736 CDEBUG(D_VFSTRACE, "VFS Op: at %llu jiffies\n", get_jiffies_64());
1737 ll_stats_ops_tally(ll_s2sbi(sb), LPROC_LL_STAFS, 1);
1739 /* Some amount of caching on the client is allowed */
1740 rc = ll_statfs_internal(sb, &osfs,
1741 cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS),
1746 statfs_unpack(sfs, &osfs);
1748 /* We need to downshift for all 32-bit kernels, because we can't
1749 * tell if the kernel is being called via sys_statfs64() or not.
1750 * Stop before overflowing f_bsize - in which case it is better
1751 * to just risk EOVERFLOW if caller is using old sys_statfs(). */
1752 if (sizeof(long) < 8) {
1753 while (osfs.os_blocks > ~0UL && sfs->f_bsize < 0x40000000) {
1756 osfs.os_blocks >>= 1;
1757 osfs.os_bfree >>= 1;
1758 osfs.os_bavail >>= 1;
1762 sfs->f_blocks = osfs.os_blocks;
1763 sfs->f_bfree = osfs.os_bfree;
1764 sfs->f_bavail = osfs.os_bavail;
1765 sfs->f_fsid.val[0] = (__u32)fsid;
1766 sfs->f_fsid.val[1] = (__u32)(fsid >> 32);
1770 void ll_inode_size_lock(struct inode *inode)
1772 struct ll_inode_info *lli;
1774 LASSERT(!S_ISDIR(inode->i_mode));
1776 lli = ll_i2info(inode);
1777 mutex_lock(&lli->lli_size_mutex);
1780 void ll_inode_size_unlock(struct inode *inode)
1782 struct ll_inode_info *lli;
1784 lli = ll_i2info(inode);
1785 mutex_unlock(&lli->lli_size_mutex);
1788 int ll_update_inode(struct inode *inode, struct lustre_md *md)
1790 struct ll_inode_info *lli = ll_i2info(inode);
1791 struct mdt_body *body = md->body;
1792 struct ll_sb_info *sbi = ll_i2sbi(inode);
1794 if (body->mbo_valid & OBD_MD_FLEASIZE)
1795 cl_file_inode_init(inode, md);
1797 if (S_ISDIR(inode->i_mode)) {
1800 rc = ll_update_lsm_md(inode, md);
1805 #ifdef CONFIG_FS_POSIX_ACL
1806 if (body->mbo_valid & OBD_MD_FLACL) {
1807 spin_lock(&lli->lli_lock);
1808 if (lli->lli_posix_acl)
1809 posix_acl_release(lli->lli_posix_acl);
1810 lli->lli_posix_acl = md->posix_acl;
1811 spin_unlock(&lli->lli_lock);
1814 inode->i_ino = cl_fid_build_ino(&body->mbo_fid1,
1815 sbi->ll_flags & LL_SBI_32BIT_API);
1816 inode->i_generation = cl_fid_build_gen(&body->mbo_fid1);
1818 if (body->mbo_valid & OBD_MD_FLATIME) {
1819 if (body->mbo_atime > LTIME_S(inode->i_atime))
1820 LTIME_S(inode->i_atime) = body->mbo_atime;
1821 lli->lli_atime = body->mbo_atime;
1824 if (body->mbo_valid & OBD_MD_FLMTIME) {
1825 if (body->mbo_mtime > LTIME_S(inode->i_mtime)) {
1826 CDEBUG(D_INODE, "setting ino %lu mtime from %lu "
1827 "to %llu\n", inode->i_ino,
1828 LTIME_S(inode->i_mtime), body->mbo_mtime);
1829 LTIME_S(inode->i_mtime) = body->mbo_mtime;
1831 lli->lli_mtime = body->mbo_mtime;
1834 if (body->mbo_valid & OBD_MD_FLCTIME) {
1835 if (body->mbo_ctime > LTIME_S(inode->i_ctime))
1836 LTIME_S(inode->i_ctime) = body->mbo_ctime;
1837 lli->lli_ctime = body->mbo_ctime;
1840 if (body->mbo_valid & OBD_MD_FLMODE)
1841 inode->i_mode = (inode->i_mode & S_IFMT) |
1842 (body->mbo_mode & ~S_IFMT);
1844 if (body->mbo_valid & OBD_MD_FLTYPE)
1845 inode->i_mode = (inode->i_mode & ~S_IFMT) |
1846 (body->mbo_mode & S_IFMT);
1848 LASSERT(inode->i_mode != 0);
1849 if (S_ISREG(inode->i_mode))
1850 inode->i_blkbits = min(PTLRPC_MAX_BRW_BITS + 1,
1851 LL_MAX_BLKSIZE_BITS);
1853 inode->i_blkbits = inode->i_sb->s_blocksize_bits;
1855 if (body->mbo_valid & OBD_MD_FLUID)
1856 inode->i_uid = make_kuid(&init_user_ns, body->mbo_uid);
1857 if (body->mbo_valid & OBD_MD_FLGID)
1858 inode->i_gid = make_kgid(&init_user_ns, body->mbo_gid);
1859 if (body->mbo_valid & OBD_MD_FLPROJID)
1860 lli->lli_projid = body->mbo_projid;
1861 if (body->mbo_valid & OBD_MD_FLFLAGS)
1862 inode->i_flags = ll_ext_to_inode_flags(body->mbo_flags);
1863 if (body->mbo_valid & OBD_MD_FLNLINK)
1864 set_nlink(inode, body->mbo_nlink);
1865 if (body->mbo_valid & OBD_MD_FLRDEV)
1866 inode->i_rdev = old_decode_dev(body->mbo_rdev);
1868 if (body->mbo_valid & OBD_MD_FLID) {
1869 /* FID shouldn't be changed! */
1870 if (fid_is_sane(&lli->lli_fid)) {
1871 LASSERTF(lu_fid_eq(&lli->lli_fid, &body->mbo_fid1),
1872 "Trying to change FID "DFID
1873 " to the "DFID", inode "DFID"(%p)\n",
1874 PFID(&lli->lli_fid), PFID(&body->mbo_fid1),
1875 PFID(ll_inode2fid(inode)), inode);
1877 lli->lli_fid = body->mbo_fid1;
1881 LASSERT(fid_seq(&lli->lli_fid) != 0);
1883 if (body->mbo_valid & OBD_MD_FLSIZE) {
1884 i_size_write(inode, body->mbo_size);
1886 CDEBUG(D_VFSTRACE, "inode="DFID", updating i_size %llu\n",
1887 PFID(ll_inode2fid(inode)),
1888 (unsigned long long)body->mbo_size);
1890 if (body->mbo_valid & OBD_MD_FLBLOCKS)
1891 inode->i_blocks = body->mbo_blocks;
1894 if (body->mbo_valid & OBD_MD_TSTATE) {
1895 /* Set LLIF_FILE_RESTORING if restore ongoing and
1896 * clear it when done to ensure to start again
1897 * glimpsing updated attrs
1899 if (body->mbo_t_state & MS_RESTORE)
1900 ll_file_set_flag(lli, LLIF_FILE_RESTORING);
1902 ll_file_clear_flag(lli, LLIF_FILE_RESTORING);
1908 int ll_read_inode2(struct inode *inode, void *opaque)
1910 struct lustre_md *md = opaque;
1911 struct ll_inode_info *lli = ll_i2info(inode);
1915 CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p)\n",
1916 PFID(&lli->lli_fid), inode);
1918 /* Core attributes from the MDS first. This is a new inode, and
1919 * the VFS doesn't zero times in the core inode so we have to do
1920 * it ourselves. They will be overwritten by either MDS or OST
1921 * attributes - we just need to make sure they aren't newer. */
1922 LTIME_S(inode->i_mtime) = 0;
1923 LTIME_S(inode->i_atime) = 0;
1924 LTIME_S(inode->i_ctime) = 0;
1926 rc = ll_update_inode(inode, md);
1930 /* OIDEBUG(inode); */
1932 #ifdef HAVE_BACKING_DEV_INFO
1933 /* initializing backing dev info. */
1934 inode->i_mapping->backing_dev_info = &s2lsi(inode->i_sb)->lsi_bdi;
1936 if (S_ISREG(inode->i_mode)) {
1937 struct ll_sb_info *sbi = ll_i2sbi(inode);
1938 inode->i_op = &ll_file_inode_operations;
1939 inode->i_fop = sbi->ll_fop;
1940 inode->i_mapping->a_ops = (struct address_space_operations *)&ll_aops;
1942 } else if (S_ISDIR(inode->i_mode)) {
1943 inode->i_op = &ll_dir_inode_operations;
1944 inode->i_fop = &ll_dir_operations;
1946 } else if (S_ISLNK(inode->i_mode)) {
1947 inode->i_op = &ll_fast_symlink_inode_operations;
1950 inode->i_op = &ll_special_inode_operations;
1952 init_special_inode(inode, inode->i_mode,
1961 void ll_delete_inode(struct inode *inode)
1963 struct ll_inode_info *lli = ll_i2info(inode);
1966 if (S_ISREG(inode->i_mode) && lli->lli_clob != NULL)
1967 /* It is last chance to write out dirty pages,
1968 * otherwise we may lose data while umount */
1969 cl_sync_file_range(inode, 0, OBD_OBJECT_EOF, CL_FSYNC_LOCAL, 1);
1971 truncate_inode_pages_final(&inode->i_data);
1973 LASSERTF(inode->i_data.nrpages == 0, "inode="DFID"(%p) nrpages=%lu, "
1974 "see https://jira.hpdd.intel.com/browse/LU-118\n",
1975 PFID(ll_inode2fid(inode)), inode, inode->i_data.nrpages);
1977 #ifdef HAVE_SBOPS_EVICT_INODE
1978 ll_clear_inode(inode);
1985 int ll_iocontrol(struct inode *inode, struct file *file,
1986 unsigned int cmd, unsigned long arg)
1988 struct ll_sb_info *sbi = ll_i2sbi(inode);
1989 struct ptlrpc_request *req = NULL;
1994 case FSFILT_IOC_GETFLAGS: {
1995 struct mdt_body *body;
1996 struct md_op_data *op_data;
1998 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL,
1999 0, 0, LUSTRE_OPC_ANY,
2001 if (IS_ERR(op_data))
2002 RETURN(PTR_ERR(op_data));
2004 op_data->op_valid = OBD_MD_FLFLAGS;
2005 rc = md_getattr(sbi->ll_md_exp, op_data, &req);
2006 ll_finish_md_op_data(op_data);
2008 CERROR("%s: failure inode "DFID": rc = %d\n",
2009 sbi->ll_md_exp->exp_obd->obd_name,
2010 PFID(ll_inode2fid(inode)), rc);
2014 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
2016 flags = body->mbo_flags;
2018 ptlrpc_req_finished(req);
2020 RETURN(put_user(flags, (int __user *)arg));
2022 case FSFILT_IOC_SETFLAGS: {
2024 struct md_op_data *op_data;
2025 struct cl_object *obj;
2027 if (get_user(flags, (int __user *)arg))
2030 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
2031 LUSTRE_OPC_ANY, NULL);
2032 if (IS_ERR(op_data))
2033 RETURN(PTR_ERR(op_data));
2035 op_data->op_attr_flags = flags;
2036 op_data->op_attr.ia_valid |= ATTR_ATTR_FLAG;
2037 rc = md_setattr(sbi->ll_md_exp, op_data, NULL, 0, &req);
2038 ll_finish_md_op_data(op_data);
2039 ptlrpc_req_finished(req);
2043 inode->i_flags = ll_ext_to_inode_flags(flags);
2045 obj = ll_i2info(inode)->lli_clob;
2049 OBD_ALLOC_PTR(attr);
2053 attr->ia_valid = ATTR_ATTR_FLAG;
2054 rc = cl_setattr_ost(obj, attr, flags);
2066 int ll_flush_ctx(struct inode *inode)
2068 struct ll_sb_info *sbi = ll_i2sbi(inode);
2070 CDEBUG(D_SEC, "flush context for user %d\n",
2071 from_kuid(&init_user_ns, current_uid()));
2073 obd_set_info_async(NULL, sbi->ll_md_exp,
2074 sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
2076 obd_set_info_async(NULL, sbi->ll_dt_exp,
2077 sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
2082 /* umount -f client means force down, don't save state */
2083 void ll_umount_begin(struct super_block *sb)
2085 struct ll_sb_info *sbi = ll_s2sbi(sb);
2086 struct obd_device *obd;
2087 struct obd_ioctl_data *ioc_data;
2088 struct l_wait_info lwi;
2089 wait_queue_head_t waitq;
2092 CDEBUG(D_VFSTRACE, "VFS Op: superblock %p count %d active %d\n", sb,
2093 sb->s_count, atomic_read(&sb->s_active));
2095 obd = class_exp2obd(sbi->ll_md_exp);
2097 CERROR("Invalid MDC connection handle %#llx\n",
2098 sbi->ll_md_exp->exp_handle.h_cookie);
2104 obd = class_exp2obd(sbi->ll_dt_exp);
2106 CERROR("Invalid LOV connection handle %#llx\n",
2107 sbi->ll_dt_exp->exp_handle.h_cookie);
2113 OBD_ALLOC_PTR(ioc_data);
2115 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_md_exp,
2116 sizeof *ioc_data, ioc_data, NULL);
2118 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_dt_exp,
2119 sizeof *ioc_data, ioc_data, NULL);
2121 OBD_FREE_PTR(ioc_data);
2124 /* Really, we'd like to wait until there are no requests outstanding,
2125 * and then continue. For now, we just periodically checking for vfs
2126 * to decrement mnt_cnt and hope to finish it within 10sec.
2128 init_waitqueue_head(&waitq);
2129 lwi = LWI_TIMEOUT_INTERVAL(cfs_time_seconds(10),
2130 cfs_time_seconds(1), NULL, NULL);
2131 l_wait_event(waitq, may_umount(sbi->ll_mnt.mnt), &lwi);
2136 int ll_remount_fs(struct super_block *sb, int *flags, char *data)
2138 struct ll_sb_info *sbi = ll_s2sbi(sb);
2139 char *profilenm = get_profile_name(sb);
2143 if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY)) {
2144 read_only = *flags & MS_RDONLY;
2145 err = obd_set_info_async(NULL, sbi->ll_md_exp,
2146 sizeof(KEY_READ_ONLY),
2147 KEY_READ_ONLY, sizeof(read_only),
2150 LCONSOLE_WARN("Failed to remount %s %s (%d)\n",
2151 profilenm, read_only ?
2152 "read-only" : "read-write", err);
2157 sb->s_flags |= MS_RDONLY;
2159 sb->s_flags &= ~MS_RDONLY;
2161 if (sbi->ll_flags & LL_SBI_VERBOSE)
2162 LCONSOLE_WARN("Remounted %s %s\n", profilenm,
2163 read_only ? "read-only" : "read-write");
2169 * Cleanup the open handle that is cached on MDT-side.
2171 * For open case, the client side open handling thread may hit error
2172 * after the MDT grant the open. Under such case, the client should
2173 * send close RPC to the MDT as cleanup; otherwise, the open handle
2174 * on the MDT will be leaked there until the client umount or evicted.
2176 * In further, if someone unlinked the file, because the open handle
2177 * holds the reference on such file/object, then it will block the
2178 * subsequent threads that want to locate such object via FID.
2180 * \param[in] sb super block for this file-system
2181 * \param[in] open_req pointer to the original open request
2183 void ll_open_cleanup(struct super_block *sb, struct ptlrpc_request *open_req)
2185 struct mdt_body *body;
2186 struct md_op_data *op_data;
2187 struct ptlrpc_request *close_req = NULL;
2188 struct obd_export *exp = ll_s2sbi(sb)->ll_md_exp;
2191 body = req_capsule_server_get(&open_req->rq_pill, &RMF_MDT_BODY);
2192 OBD_ALLOC_PTR(op_data);
2193 if (op_data == NULL) {
2194 CWARN("%s: cannot allocate op_data to release open handle for "
2196 ll_get_fsname(sb, NULL, 0), PFID(&body->mbo_fid1));
2201 op_data->op_fid1 = body->mbo_fid1;
2202 op_data->op_handle = body->mbo_handle;
2203 op_data->op_mod_time = ktime_get_real_seconds();
2204 md_close(exp, op_data, NULL, &close_req);
2205 ptlrpc_req_finished(close_req);
2206 ll_finish_md_op_data(op_data);
2211 int ll_prep_inode(struct inode **inode, struct ptlrpc_request *req,
2212 struct super_block *sb, struct lookup_intent *it)
2214 struct ll_sb_info *sbi = NULL;
2215 struct lustre_md md = { NULL };
2219 LASSERT(*inode || sb);
2220 sbi = sb ? ll_s2sbi(sb) : ll_i2sbi(*inode);
2221 rc = md_get_lustre_md(sbi->ll_md_exp, req, sbi->ll_dt_exp,
2222 sbi->ll_md_exp, &md);
2227 rc = ll_update_inode(*inode, &md);
2231 LASSERT(sb != NULL);
2234 * At this point server returns to client's same fid as client
2235 * generated for creating. So using ->fid1 is okay here.
2237 if (!fid_is_sane(&md.body->mbo_fid1)) {
2238 CERROR("%s: Fid is insane "DFID"\n",
2239 ll_get_fsname(sb, NULL, 0),
2240 PFID(&md.body->mbo_fid1));
2241 GOTO(out, rc = -EINVAL);
2244 *inode = ll_iget(sb, cl_fid_build_ino(&md.body->mbo_fid1,
2245 sbi->ll_flags & LL_SBI_32BIT_API),
2247 if (IS_ERR(*inode)) {
2248 #ifdef CONFIG_FS_POSIX_ACL
2250 posix_acl_release(md.posix_acl);
2251 md.posix_acl = NULL;
2254 rc = IS_ERR(*inode) ? PTR_ERR(*inode) : -ENOMEM;
2256 CERROR("new_inode -fatal: rc %d\n", rc);
2261 /* Handling piggyback layout lock.
2262 * Layout lock can be piggybacked by getattr and open request.
2263 * The lsm can be applied to inode only if it comes with a layout lock
2264 * otherwise correct layout may be overwritten, for example:
2265 * 1. proc1: mdt returns a lsm but not granting layout
2266 * 2. layout was changed by another client
2267 * 3. proc2: refresh layout and layout lock granted
2268 * 4. proc1: to apply a stale layout */
2269 if (it != NULL && it->it_lock_mode != 0) {
2270 struct lustre_handle lockh;
2271 struct ldlm_lock *lock;
2273 lockh.cookie = it->it_lock_handle;
2274 lock = ldlm_handle2lock(&lockh);
2275 LASSERT(lock != NULL);
2276 if (ldlm_has_layout(lock)) {
2277 struct cl_object_conf conf;
2279 memset(&conf, 0, sizeof(conf));
2280 conf.coc_opc = OBJECT_CONF_SET;
2281 conf.coc_inode = *inode;
2282 conf.coc_lock = lock;
2283 conf.u.coc_layout = md.layout;
2284 (void)ll_layout_conf(*inode, &conf);
2286 LDLM_LOCK_PUT(lock);
2292 md_free_lustre_md(sbi->ll_md_exp, &md);
2295 if (rc != 0 && it != NULL && it->it_op & IT_OPEN)
2296 ll_open_cleanup(sb != NULL ? sb : (*inode)->i_sb, req);
2301 int ll_obd_statfs(struct inode *inode, void __user *arg)
2303 struct ll_sb_info *sbi = NULL;
2304 struct obd_export *exp;
2306 struct obd_ioctl_data *data = NULL;
2310 if (!inode || !(sbi = ll_i2sbi(inode)))
2311 GOTO(out_statfs, rc = -EINVAL);
2313 rc = obd_ioctl_getdata(&buf, &len, arg);
2315 GOTO(out_statfs, rc);
2318 if (!data->ioc_inlbuf1 || !data->ioc_inlbuf2 ||
2319 !data->ioc_pbuf1 || !data->ioc_pbuf2)
2320 GOTO(out_statfs, rc = -EINVAL);
2322 if (data->ioc_inllen1 != sizeof(__u32) ||
2323 data->ioc_inllen2 != sizeof(__u32) ||
2324 data->ioc_plen1 != sizeof(struct obd_statfs) ||
2325 data->ioc_plen2 != sizeof(struct obd_uuid))
2326 GOTO(out_statfs, rc = -EINVAL);
2328 memcpy(&type, data->ioc_inlbuf1, sizeof(__u32));
2329 if (type & LL_STATFS_LMV)
2330 exp = sbi->ll_md_exp;
2331 else if (type & LL_STATFS_LOV)
2332 exp = sbi->ll_dt_exp;
2334 GOTO(out_statfs, rc = -ENODEV);
2336 rc = obd_iocontrol(IOC_OBD_STATFS, exp, len, buf, NULL);
2338 GOTO(out_statfs, rc);
2340 OBD_FREE_LARGE(buf, len);
2344 int ll_process_config(struct lustre_cfg *lcfg)
2346 struct super_block *sb;
2351 /* The instance name contains the sb: lustre-client-aacfe000 */
2352 ptr = strrchr(lustre_cfg_string(lcfg, 0), '-');
2353 if (!ptr || !*(++ptr))
2355 if (sscanf(ptr, "%lx", &x) != 1)
2357 sb = (struct super_block *)x;
2358 /* This better be a real Lustre superblock! */
2359 LASSERT(s2lsi(sb)->lsi_lmd->lmd_magic == LMD_MAGIC);
2361 /* Note we have not called client_common_fill_super yet, so
2362 proc fns must be able to handle that! */
2363 rc = class_process_proc_param(PARAM_LLITE, lprocfs_llite_obd_vars,
2370 /* this function prepares md_op_data hint for passing it down to MD stack. */
2371 struct md_op_data *ll_prep_md_op_data(struct md_op_data *op_data,
2372 struct inode *i1, struct inode *i2,
2373 const char *name, size_t namelen,
2374 __u32 mode, __u32 opc, void *data)
2376 LASSERT(i1 != NULL);
2379 /* Do not reuse namelen for something else. */
2381 return ERR_PTR(-EINVAL);
2383 if (namelen > ll_i2sbi(i1)->ll_namelen)
2384 return ERR_PTR(-ENAMETOOLONG);
2386 if (!lu_name_is_valid_2(name, namelen))
2387 return ERR_PTR(-EINVAL);
2390 if (op_data == NULL)
2391 OBD_ALLOC_PTR(op_data);
2393 if (op_data == NULL)
2394 return ERR_PTR(-ENOMEM);
2396 ll_i2gids(op_data->op_suppgids, i1, i2);
2397 op_data->op_fid1 = *ll_inode2fid(i1);
2398 op_data->op_default_stripe_offset = -1;
2399 if (S_ISDIR(i1->i_mode)) {
2400 op_data->op_mea1 = ll_i2info(i1)->lli_lsm_md;
2401 if (opc == LUSTRE_OPC_MKDIR)
2402 op_data->op_default_stripe_offset =
2403 ll_i2info(i1)->lli_def_stripe_offset;
2407 op_data->op_fid2 = *ll_inode2fid(i2);
2408 if (S_ISDIR(i2->i_mode))
2409 op_data->op_mea2 = ll_i2info(i2)->lli_lsm_md;
2411 fid_zero(&op_data->op_fid2);
2414 if (ll_i2sbi(i1)->ll_flags & LL_SBI_64BIT_HASH)
2415 op_data->op_cli_flags |= CLI_HASH64;
2417 if (ll_need_32bit_api(ll_i2sbi(i1)))
2418 op_data->op_cli_flags |= CLI_API32;
2420 op_data->op_name = name;
2421 op_data->op_namelen = namelen;
2422 op_data->op_mode = mode;
2423 op_data->op_mod_time = cfs_time_current_sec();
2424 op_data->op_fsuid = from_kuid(&init_user_ns, current_fsuid());
2425 op_data->op_fsgid = from_kgid(&init_user_ns, current_fsgid());
2426 op_data->op_cap = cfs_curproc_cap_pack();
2427 if ((opc == LUSTRE_OPC_CREATE) && (name != NULL) &&
2428 filename_is_volatile(name, namelen, &op_data->op_mds)) {
2429 op_data->op_bias |= MDS_CREATE_VOLATILE;
2431 op_data->op_mds = 0;
2433 op_data->op_data = data;
2438 void ll_finish_md_op_data(struct md_op_data *op_data)
2440 security_release_secctx(op_data->op_file_secctx,
2441 op_data->op_file_secctx_size);
2442 OBD_FREE_PTR(op_data);
2445 #ifdef HAVE_SUPEROPS_USE_DENTRY
2446 int ll_show_options(struct seq_file *seq, struct dentry *dentry)
2448 int ll_show_options(struct seq_file *seq, struct vfsmount *vfs)
2451 struct ll_sb_info *sbi;
2453 #ifdef HAVE_SUPEROPS_USE_DENTRY
2454 LASSERT((seq != NULL) && (dentry != NULL));
2455 sbi = ll_s2sbi(dentry->d_sb);
2457 LASSERT((seq != NULL) && (vfs != NULL));
2458 sbi = ll_s2sbi(vfs->mnt_sb);
2461 if (sbi->ll_flags & LL_SBI_NOLCK)
2462 seq_puts(seq, ",nolock");
2464 if (sbi->ll_flags & LL_SBI_FLOCK)
2465 seq_puts(seq, ",flock");
2467 if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
2468 seq_puts(seq, ",localflock");
2470 if (sbi->ll_flags & LL_SBI_USER_XATTR)
2471 seq_puts(seq, ",user_xattr");
2473 if (sbi->ll_flags & LL_SBI_LAZYSTATFS)
2474 seq_puts(seq, ",lazystatfs");
2476 if (sbi->ll_flags & LL_SBI_USER_FID2PATH)
2477 seq_puts(seq, ",user_fid2path");
2479 if (sbi->ll_flags & LL_SBI_ALWAYS_PING)
2480 seq_puts(seq, ",always_ping");
2486 * Get obd name by cmd, and copy out to user space
2488 int ll_get_obd_name(struct inode *inode, unsigned int cmd, unsigned long arg)
2490 struct ll_sb_info *sbi = ll_i2sbi(inode);
2491 struct obd_device *obd;
2494 if (cmd == OBD_IOC_GETDTNAME)
2495 obd = class_exp2obd(sbi->ll_dt_exp);
2496 else if (cmd == OBD_IOC_GETMDNAME)
2497 obd = class_exp2obd(sbi->ll_md_exp);
2504 if (copy_to_user((void __user *)arg, obd->obd_name,
2505 strlen(obd->obd_name) + 1))
2512 * Get lustre file system name by \a sbi. If \a buf is provided(non-NULL), the
2513 * fsname will be returned in this buffer; otherwise, a static buffer will be
2514 * used to store the fsname and returned to caller.
2516 char *ll_get_fsname(struct super_block *sb, char *buf, int buflen)
2518 static char fsname_static[MTI_NAME_MAXLEN];
2519 struct lustre_sb_info *lsi = s2lsi(sb);
2524 /* this means the caller wants to use static buffer
2525 * and it doesn't care about race. Usually this is
2526 * in error reporting path */
2527 buf = fsname_static;
2528 buflen = sizeof(fsname_static);
2531 len = strlen(lsi->lsi_lmd->lmd_profile);
2532 ptr = strrchr(lsi->lsi_lmd->lmd_profile, '-');
2533 if (ptr && (strcmp(ptr, "-client") == 0))
2536 if (unlikely(len >= buflen))
2538 strncpy(buf, lsi->lsi_lmd->lmd_profile, len);
2544 static char* ll_d_path(struct dentry *dentry, char *buf, int bufsize)
2551 p.mnt = current->fs->root.mnt;
2553 path = d_path(&p, buf, bufsize);
2558 void ll_dirty_page_discard_warn(struct page *page, int ioret)
2560 char *buf, *path = NULL;
2561 struct dentry *dentry = NULL;
2562 struct inode *inode = page->mapping->host;
2564 /* this can be called inside spin lock so use GFP_ATOMIC. */
2565 buf = (char *)__get_free_page(GFP_ATOMIC);
2567 dentry = d_find_alias(page->mapping->host);
2569 path = ll_d_path(dentry, buf, PAGE_SIZE);
2573 "%s: dirty page discard: %s/fid: "DFID"/%s may get corrupted "
2574 "(rc %d)\n", ll_get_fsname(page->mapping->host->i_sb, NULL, 0),
2575 s2lsi(page->mapping->host->i_sb)->lsi_lmd->lmd_dev,
2576 PFID(ll_inode2fid(inode)),
2577 (path && !IS_ERR(path)) ? path : "", ioret);
2583 free_page((unsigned long)buf);
2586 ssize_t ll_copy_user_md(const struct lov_user_md __user *md,
2587 struct lov_user_md **kbuf)
2589 struct lov_user_md lum;
2593 if (copy_from_user(&lum, md, sizeof(lum)))
2596 lum_size = ll_lov_user_md_size(&lum);
2600 OBD_ALLOC(*kbuf, lum_size);
2604 if (copy_from_user(*kbuf, md, lum_size) != 0) {
2605 OBD_FREE(*kbuf, lum_size);
2613 * Compute llite root squash state after a change of root squash
2614 * configuration setting or add/remove of a lnet nid
2616 void ll_compute_rootsquash_state(struct ll_sb_info *sbi)
2618 struct root_squash_info *squash = &sbi->ll_squash;
2621 struct lnet_process_id id;
2623 /* Update norootsquash flag */
2624 down_write(&squash->rsi_sem);
2625 if (list_empty(&squash->rsi_nosquash_nids))
2626 sbi->ll_flags &= ~LL_SBI_NOROOTSQUASH;
2628 /* Do not apply root squash as soon as one of our NIDs is
2629 * in the nosquash_nids list */
2632 while (LNetGetId(i++, &id) != -ENOENT) {
2633 if (LNET_NETTYP(LNET_NIDNET(id.nid)) == LOLND)
2635 if (cfs_match_nid(id.nid, &squash->rsi_nosquash_nids)) {
2641 sbi->ll_flags |= LL_SBI_NOROOTSQUASH;
2643 sbi->ll_flags &= ~LL_SBI_NOROOTSQUASH;
2645 up_write(&squash->rsi_sem);
2649 * Parse linkea content to extract information about a given hardlink
2651 * \param[in] ldata - Initialized linkea data
2652 * \param[in] linkno - Link identifier
2653 * \param[out] parent_fid - The entry's parent FID
2654 * \param[out] ln - Entry name destination buffer
2656 * \retval 0 on success
2657 * \retval Appropriate negative error code on failure
2659 static int ll_linkea_decode(struct linkea_data *ldata, unsigned int linkno,
2660 struct lu_fid *parent_fid, struct lu_name *ln)
2666 rc = linkea_init_with_rec(ldata);
2670 if (linkno >= ldata->ld_leh->leh_reccount)
2671 /* beyond last link */
2674 linkea_first_entry(ldata);
2675 for (idx = 0; ldata->ld_lee != NULL; idx++) {
2676 linkea_entry_unpack(ldata->ld_lee, &ldata->ld_reclen, ln,
2681 linkea_next_entry(ldata);
2691 * Get parent FID and name of an identified link. Operation is performed for
2692 * a given link number, letting the caller iterate over linkno to list one or
2693 * all links of an entry.
2695 * \param[in] file - File descriptor against which to perform the operation
2696 * \param[in,out] arg - User-filled structure containing the linkno to operate
2697 * on and the available size. It is eventually filled with
2698 * the requested information or left untouched on error
2700 * \retval - 0 on success
2701 * \retval - Appropriate negative error code on failure
2703 int ll_getparent(struct file *file, struct getparent __user *arg)
2705 struct dentry *dentry = file_dentry(file);
2706 struct inode *inode = file_inode(file);
2707 struct linkea_data *ldata;
2708 struct lu_buf buf = LU_BUF_NULL;
2710 struct lu_fid parent_fid;
2717 if (!cfs_capable(CFS_CAP_DAC_READ_SEARCH) &&
2718 !(ll_i2sbi(inode)->ll_flags & LL_SBI_USER_FID2PATH))
2721 if (get_user(name_size, &arg->gp_name_size))
2724 if (get_user(linkno, &arg->gp_linkno))
2727 if (name_size > PATH_MAX)
2730 OBD_ALLOC(ldata, sizeof(*ldata));
2734 rc = linkea_data_new(ldata, &buf);
2736 GOTO(ldata_free, rc);
2738 rc = ll_getxattr(dentry, XATTR_NAME_LINK, buf.lb_buf, buf.lb_len);
2742 rc = ll_linkea_decode(ldata, linkno, &parent_fid, &ln);
2746 if (ln.ln_namelen >= name_size)
2747 GOTO(lb_free, rc = -EOVERFLOW);
2749 if (copy_to_user(&arg->gp_fid, &parent_fid, sizeof(arg->gp_fid)))
2750 GOTO(lb_free, rc = -EFAULT);
2752 if (copy_to_user(&arg->gp_name, ln.ln_name, ln.ln_namelen))
2753 GOTO(lb_free, rc = -EFAULT);
2755 if (put_user('\0', arg->gp_name + ln.ln_namelen))
2756 GOTO(lb_free, rc = -EFAULT);
2761 OBD_FREE(ldata, sizeof(*ldata));