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.sun.com/software/products/lustre/docs/GPLv2.pdf
20 * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
21 * CA 95054 USA or visit www.sun.com if you need additional information or
27 * Copyright (c) 2003, 2010, Oracle and/or its affiliates. All rights reserved.
28 * Use is subject to license terms.
30 * Copyright (c) 2011, 2012, Intel Corporation.
33 * This file is part of Lustre, http://www.lustre.org/
34 * Lustre is a trademark of Sun Microsystems, Inc.
36 * lustre/llite/llite_lib.c
38 * Lustre Light Super operations
41 #define DEBUG_SUBSYSTEM S_LLITE
43 #include <linux/module.h>
44 #include <linux/types.h>
47 #include <lustre_lite.h>
48 #include <lustre_ha.h>
49 #include <lustre_dlm.h>
50 #include <lprocfs_status.h>
51 #include <lustre_disk.h>
52 #include <lustre_param.h>
53 #include <lustre_log.h>
54 #include <cl_object.h>
55 #include <obd_cksum.h>
56 #include "llite_internal.h"
58 struct kmem_cache *ll_file_data_slab;
59 struct proc_dir_entry *proc_lustre_fs_root;
61 LIST_HEAD(ll_super_blocks);
62 DEFINE_SPINLOCK(ll_sb_lock);
64 #ifndef MS_HAS_NEW_AOPS
65 extern struct address_space_operations ll_aops;
67 extern struct address_space_operations_ext ll_aops;
71 #define log2(n) ffz(~(n))
74 static struct ll_sb_info *ll_init_sbi(void)
76 struct ll_sb_info *sbi = NULL;
78 unsigned long lru_page_max;
83 OBD_ALLOC(sbi, sizeof(*sbi));
87 spin_lock_init(&sbi->ll_lock);
88 mutex_init(&sbi->ll_lco.lco_lock);
89 spin_lock_init(&sbi->ll_pp_extent_lock);
90 spin_lock_init(&sbi->ll_process_lock);
91 sbi->ll_rw_stats_on = 0;
94 pages = si.totalram - si.totalhigh;
95 if (pages >> (20 - PAGE_CACHE_SHIFT) < 512) {
96 lru_page_max = pages / 2;
98 lru_page_max = (pages / 4) * 3;
101 /* initialize lru data */
102 atomic_set(&sbi->ll_cache.ccc_users, 0);
103 sbi->ll_cache.ccc_lru_max = lru_page_max;
104 atomic_set(&sbi->ll_cache.ccc_lru_left, lru_page_max);
105 spin_lock_init(&sbi->ll_cache.ccc_lru_lock);
106 INIT_LIST_HEAD(&sbi->ll_cache.ccc_lru);
108 sbi->ll_ra_info.ra_max_pages_per_file = min(pages / 32,
109 SBI_DEFAULT_READAHEAD_MAX);
110 sbi->ll_ra_info.ra_max_pages = sbi->ll_ra_info.ra_max_pages_per_file;
111 sbi->ll_ra_info.ra_max_read_ahead_whole_pages =
112 SBI_DEFAULT_READAHEAD_WHOLE_MAX;
113 INIT_LIST_HEAD(&sbi->ll_conn_chain);
114 INIT_LIST_HEAD(&sbi->ll_orphan_dentry_list);
116 ll_generate_random_uuid(uuid);
117 class_uuid_unparse(uuid, &sbi->ll_sb_uuid);
118 CDEBUG(D_CONFIG, "generated uuid: %s\n", sbi->ll_sb_uuid.uuid);
120 spin_lock(&ll_sb_lock);
121 list_add_tail(&sbi->ll_list, &ll_super_blocks);
122 spin_unlock(&ll_sb_lock);
124 sbi->ll_flags |= LL_SBI_VERBOSE;
125 sbi->ll_flags |= LL_SBI_CHECKSUM;
127 sbi->ll_flags |= LL_SBI_LRU_RESIZE;
129 for (i = 0; i <= LL_PROCESS_HIST_MAX; i++) {
130 spin_lock_init(&sbi->ll_rw_extents_info.pp_extents[i].
132 spin_lock_init(&sbi->ll_rw_extents_info.pp_extents[i].
136 /* metadata statahead is enabled by default */
137 sbi->ll_sa_max = LL_SA_RPC_DEF;
138 atomic_set(&sbi->ll_sa_total, 0);
139 atomic_set(&sbi->ll_sa_wrong, 0);
140 atomic_set(&sbi->ll_agl_total, 0);
141 sbi->ll_flags |= LL_SBI_AGL_ENABLED;
146 void ll_free_sbi(struct super_block *sb)
148 struct ll_sb_info *sbi = ll_s2sbi(sb);
151 spin_lock(&ll_sb_lock);
152 list_del(&sbi->ll_list);
153 spin_unlock(&ll_sb_lock);
154 OBD_FREE(sbi, sizeof(*sbi));
158 static struct dentry_operations ll_d_root_ops = {
159 .d_compare = ll_dcompare,
160 .d_revalidate = ll_revalidate_nd,
163 static int client_common_fill_super(struct super_block *sb, char *md, char *dt,
164 struct vfsmount *mnt)
166 struct inode *root = 0;
167 struct ll_sb_info *sbi = ll_s2sbi(sb);
168 struct obd_device *obd;
169 struct obd_capa *oc = NULL;
170 struct obd_statfs *osfs = NULL;
171 struct ptlrpc_request *request = NULL;
172 struct obd_connect_data *data = NULL;
173 struct obd_uuid *uuid;
174 struct md_op_data *op_data;
175 struct lustre_md lmd;
177 int size, err, checksum;
179 obd = class_name2obd(md);
181 CERROR("MD %s: not setup or attached\n", md);
195 if (proc_lustre_fs_root) {
196 err = lprocfs_register_mountpoint(proc_lustre_fs_root, sb,
199 CERROR("could not register mount in /proc/fs/lustre\n");
202 /* indicate the features supported by this client */
203 data->ocd_connect_flags = OBD_CONNECT_IBITS | OBD_CONNECT_NODEVOH |
204 OBD_CONNECT_ATTRFID |
205 OBD_CONNECT_VERSION | OBD_CONNECT_BRW_SIZE |
206 OBD_CONNECT_MDS_CAPA | OBD_CONNECT_OSS_CAPA |
207 OBD_CONNECT_CANCELSET | OBD_CONNECT_FID |
208 OBD_CONNECT_AT | OBD_CONNECT_LOV_V3 |
209 OBD_CONNECT_RMT_CLIENT | OBD_CONNECT_VBR |
210 OBD_CONNECT_FULL20 | OBD_CONNECT_64BITHASH|
211 OBD_CONNECT_EINPROGRESS |
212 OBD_CONNECT_JOBSTATS | OBD_CONNECT_LVB_TYPE |
213 OBD_CONNECT_LAYOUTLOCK |
214 OBD_CONNECT_PINGLESS | OBD_CONNECT_MAX_EASIZE;
216 if (sbi->ll_flags & LL_SBI_SOM_PREVIEW)
217 data->ocd_connect_flags |= OBD_CONNECT_SOM;
219 if (sbi->ll_flags & LL_SBI_LRU_RESIZE)
220 data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
221 #ifdef CONFIG_FS_POSIX_ACL
222 data->ocd_connect_flags |= OBD_CONNECT_ACL | OBD_CONNECT_UMASK;
225 if (OBD_FAIL_CHECK(OBD_FAIL_MDC_LIGHTWEIGHT))
226 /* flag mdc connection as lightweight, only used for test
227 * purpose, use with care */
228 data->ocd_connect_flags |= OBD_CONNECT_LIGHTWEIGHT;
230 data->ocd_ibits_known = MDS_INODELOCK_FULL;
231 data->ocd_version = LUSTRE_VERSION_CODE;
233 if (sb->s_flags & MS_RDONLY)
234 data->ocd_connect_flags |= OBD_CONNECT_RDONLY;
235 if (sbi->ll_flags & LL_SBI_USER_XATTR)
236 data->ocd_connect_flags |= OBD_CONNECT_XATTR;
238 #ifdef HAVE_MS_FLOCK_LOCK
239 /* force vfs to use lustre handler for flock() calls - bug 10743 */
240 sb->s_flags |= MS_FLOCK_LOCK;
242 #ifdef MS_HAS_NEW_AOPS
243 sb->s_flags |= MS_HAS_NEW_AOPS;
246 if (sbi->ll_flags & LL_SBI_FLOCK)
247 sbi->ll_fop = &ll_file_operations_flock;
248 else if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
249 sbi->ll_fop = &ll_file_operations;
251 sbi->ll_fop = &ll_file_operations_noflock;
254 data->ocd_connect_flags |= OBD_CONNECT_REAL;
255 if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
256 data->ocd_connect_flags |= OBD_CONNECT_RMT_CLIENT_FORCE;
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_CACHE_SIZE);
304 obd_connect_flags2str(buf, PAGE_CACHE_SIZE,
305 valid ^ CLIENT_CONNECT_MDT_REQD, ",");
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_CACHE_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, NULL);
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_max_rw_chunk = LL_DEFAULT_MAX_RW_CHUNK;
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_RMT_CLIENT) {
353 if (!(sbi->ll_flags & LL_SBI_RMT_CLIENT)) {
354 sbi->ll_flags |= LL_SBI_RMT_CLIENT;
355 LCONSOLE_INFO("client is set as remote by default.\n");
358 if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
359 sbi->ll_flags &= ~LL_SBI_RMT_CLIENT;
360 LCONSOLE_INFO("client claims to be remote, but server "
361 "rejected, forced to be local.\n");
365 if (data->ocd_connect_flags & OBD_CONNECT_MDS_CAPA) {
366 LCONSOLE_INFO("client enabled MDS capability!\n");
367 sbi->ll_flags |= LL_SBI_MDS_CAPA;
370 if (data->ocd_connect_flags & OBD_CONNECT_OSS_CAPA) {
371 LCONSOLE_INFO("client enabled OSS capability!\n");
372 sbi->ll_flags |= LL_SBI_OSS_CAPA;
375 if (data->ocd_connect_flags & OBD_CONNECT_64BITHASH)
376 sbi->ll_flags |= LL_SBI_64BIT_HASH;
378 if (data->ocd_connect_flags & OBD_CONNECT_BRW_SIZE)
379 sbi->ll_md_brw_size = data->ocd_brw_size;
381 sbi->ll_md_brw_size = PAGE_CACHE_SIZE;
383 if (data->ocd_connect_flags & OBD_CONNECT_LAYOUTLOCK) {
384 LCONSOLE_INFO("Layout lock feature supported.\n");
385 sbi->ll_flags |= LL_SBI_LAYOUT_LOCK;
388 if (data->ocd_ibits_known & MDS_INODELOCK_XATTR) {
389 if (!(data->ocd_connect_flags & OBD_CONNECT_MAX_EASIZE)) {
391 "%s: disabling xattr cache due to unknown maximum xattr size.\n",
394 sbi->ll_flags |= LL_SBI_XATTR_CACHE;
395 sbi->ll_xattr_cache_enabled = 1;
399 obd = class_name2obd(dt);
401 CERROR("DT %s: not setup or attached\n", dt);
402 GOTO(out_md_fid, err = -ENODEV);
405 data->ocd_connect_flags = OBD_CONNECT_GRANT | OBD_CONNECT_VERSION |
406 OBD_CONNECT_REQPORTAL | OBD_CONNECT_BRW_SIZE |
407 OBD_CONNECT_CANCELSET | OBD_CONNECT_FID |
408 OBD_CONNECT_SRVLOCK | OBD_CONNECT_TRUNCLOCK|
409 OBD_CONNECT_AT | OBD_CONNECT_RMT_CLIENT |
410 OBD_CONNECT_OSS_CAPA | OBD_CONNECT_VBR|
411 OBD_CONNECT_FULL20 | OBD_CONNECT_64BITHASH |
412 OBD_CONNECT_MAXBYTES |
413 OBD_CONNECT_EINPROGRESS |
414 OBD_CONNECT_JOBSTATS | OBD_CONNECT_LVB_TYPE |
415 OBD_CONNECT_LAYOUTLOCK | OBD_CONNECT_PINGLESS;
417 if (sbi->ll_flags & LL_SBI_SOM_PREVIEW)
418 data->ocd_connect_flags |= OBD_CONNECT_SOM;
420 if (!OBD_FAIL_CHECK(OBD_FAIL_OSC_CONNECT_CKSUM)) {
421 /* OBD_CONNECT_CKSUM should always be set, even if checksums are
422 * disabled by default, because it can still be enabled on the
423 * fly via /proc. As a consequence, we still need to come to an
424 * agreement on the supported algorithms at connect time */
425 data->ocd_connect_flags |= OBD_CONNECT_CKSUM;
427 if (OBD_FAIL_CHECK(OBD_FAIL_OSC_CKSUM_ADLER_ONLY))
428 data->ocd_cksum_types = OBD_CKSUM_ADLER;
430 data->ocd_cksum_types = cksum_types_supported_client();
433 data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
434 if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
435 data->ocd_connect_flags |= OBD_CONNECT_RMT_CLIENT_FORCE;
437 CDEBUG(D_RPCTRACE, "ocd_connect_flags: "LPX64" ocd_version: %d "
438 "ocd_grant: %d\n", data->ocd_connect_flags,
439 data->ocd_version, data->ocd_grant);
441 obd->obd_upcall.onu_owner = &sbi->ll_lco;
442 obd->obd_upcall.onu_upcall = cl_ocd_update;
444 data->ocd_brw_size = DT_MAX_BRW_SIZE;
446 err = obd_connect(NULL, &sbi->ll_dt_exp, obd, &sbi->ll_sb_uuid, data,
449 LCONSOLE_ERROR_MSG(0x150, "An OST (dt %s) is performing "
450 "recovery, of which this client is not a "
451 "part. Please wait for recovery to "
452 "complete, abort, or time out.\n", dt);
455 CERROR("%s: Cannot connect to %s: rc = %d\n",
456 sbi->ll_dt_exp->exp_obd->obd_name, dt, err);
460 sbi->ll_dt_exp->exp_connect_data = *data;
462 err = obd_fid_init(sbi->ll_dt_exp->exp_obd, sbi->ll_dt_exp,
463 LUSTRE_SEQ_METADATA);
465 CERROR("%s: Can't init data layer FID infrastructure, "
466 "rc = %d\n", sbi->ll_dt_exp->exp_obd->obd_name, err);
470 mutex_lock(&sbi->ll_lco.lco_lock);
471 sbi->ll_lco.lco_flags = data->ocd_connect_flags;
472 sbi->ll_lco.lco_md_exp = sbi->ll_md_exp;
473 sbi->ll_lco.lco_dt_exp = sbi->ll_dt_exp;
474 mutex_unlock(&sbi->ll_lco.lco_lock);
476 fid_zero(&sbi->ll_root_fid);
477 err = md_getstatus(sbi->ll_md_exp, &sbi->ll_root_fid, &oc);
479 CERROR("cannot mds_connect: rc = %d\n", err);
480 GOTO(out_lock_cn_cb, err);
482 if (!fid_is_sane(&sbi->ll_root_fid)) {
483 CERROR("%s: Invalid root fid "DFID" during mount\n",
484 sbi->ll_md_exp->exp_obd->obd_name,
485 PFID(&sbi->ll_root_fid));
486 GOTO(out_lock_cn_cb, err = -EINVAL);
488 CDEBUG(D_SUPER, "rootfid "DFID"\n", PFID(&sbi->ll_root_fid));
490 sb->s_op = &lustre_super_operations;
491 #if THREAD_SIZE >= 8192 /*b=17630*/
492 sb->s_export_op = &lustre_export_operations;
496 * XXX: move this to after cbd setup? */
497 valid = OBD_MD_FLGETATTR | OBD_MD_FLBLOCKS | OBD_MD_FLMDSCAPA;
498 if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
499 valid |= OBD_MD_FLRMTPERM;
500 else if (sbi->ll_flags & LL_SBI_ACL)
501 valid |= OBD_MD_FLACL;
503 OBD_ALLOC_PTR(op_data);
505 GOTO(out_lock_cn_cb, err = -ENOMEM);
507 op_data->op_fid1 = sbi->ll_root_fid;
508 op_data->op_mode = 0;
509 op_data->op_capa1 = oc;
510 op_data->op_valid = valid;
512 err = md_getattr(sbi->ll_md_exp, op_data, &request);
515 OBD_FREE_PTR(op_data);
517 CERROR("%s: md_getattr failed for root: rc = %d\n",
518 sbi->ll_md_exp->exp_obd->obd_name, err);
519 GOTO(out_lock_cn_cb, err);
522 err = md_get_lustre_md(sbi->ll_md_exp, request, sbi->ll_dt_exp,
523 sbi->ll_md_exp, &lmd);
525 CERROR("failed to understand root inode md: rc = %d\n", err);
526 ptlrpc_req_finished(request);
527 GOTO(out_lock_cn_cb, err);
530 LASSERT(fid_is_sane(&sbi->ll_root_fid));
531 root = ll_iget(sb, cl_fid_build_ino(&sbi->ll_root_fid,
532 sbi->ll_flags & LL_SBI_32BIT_API),
534 md_free_lustre_md(sbi->ll_md_exp, &lmd);
535 ptlrpc_req_finished(request);
537 if (root == NULL || IS_ERR(root)) {
539 obd_free_memmd(sbi->ll_dt_exp, &lmd.lsm);
540 #ifdef CONFIG_FS_POSIX_ACL
542 posix_acl_release(lmd.posix_acl);
543 lmd.posix_acl = NULL;
546 err = IS_ERR(root) ? PTR_ERR(root) : -EBADF;
548 CERROR("lustre_lite: bad iget4 for root\n");
552 err = ll_close_thread_start(&sbi->ll_lcq);
554 CERROR("cannot start close thread: rc %d\n", err);
558 #ifdef CONFIG_FS_POSIX_ACL
559 if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
560 rct_init(&sbi->ll_rct);
561 et_init(&sbi->ll_et);
565 checksum = sbi->ll_flags & LL_SBI_CHECKSUM;
566 err = obd_set_info_async(NULL, sbi->ll_dt_exp, sizeof(KEY_CHECKSUM),
567 KEY_CHECKSUM, sizeof(checksum), &checksum,
571 err = obd_set_info_async(NULL, sbi->ll_dt_exp, sizeof(KEY_CACHE_SET),
572 KEY_CACHE_SET, sizeof(sbi->ll_cache),
573 &sbi->ll_cache, NULL);
575 sb->s_root = d_make_root(root);
576 if (sb->s_root == NULL) {
577 CERROR("%s: can't make root dentry\n",
578 ll_get_fsname(sb, NULL, 0));
579 GOTO(out_root, err = -ENOMEM);
582 /* kernel >= 2.6.38 store dentry operations in sb->s_d_op. */
583 d_set_d_op(sb->s_root, &ll_d_root_ops);
584 sb->s_d_op = &ll_d_ops;
586 sbi->ll_sdev_orig = sb->s_dev;
588 /* We set sb->s_dev equal on all lustre clients in order to support
589 * NFS export clustering. NFSD requires that the FSID be the same
591 /* s_dev is also used in lt_compare() to compare two fs, but that is
592 * only a node-local comparison. */
593 uuid = obd_get_uuid(sbi->ll_md_exp);
595 sb->s_dev = get_uuid2int(uuid->uuid, strlen(uuid->uuid));
596 get_uuid2fsid(uuid->uuid, strlen(uuid->uuid), &sbi->ll_fsid);
609 obd_fid_fini(sbi->ll_dt_exp->exp_obd);
611 obd_disconnect(sbi->ll_dt_exp);
612 sbi->ll_dt_exp = NULL;
613 /* Make sure all OScs are gone, since cl_cache is accessing sbi. */
614 obd_zombie_barrier();
616 obd_fid_fini(sbi->ll_md_exp->exp_obd);
618 obd_disconnect(sbi->ll_md_exp);
619 sbi->ll_md_exp = NULL;
625 lprocfs_unregister_mountpoint(sbi);
629 int ll_get_max_mdsize(struct ll_sb_info *sbi, int *lmmsize)
633 *lmmsize = obd_size_diskmd(sbi->ll_dt_exp, NULL);
635 rc = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_MAX_EASIZE),
636 KEY_MAX_EASIZE, &size, lmmsize, NULL);
638 CERROR("Get max mdsize error rc %d \n", rc);
643 void ll_dump_inode(struct inode *inode)
645 struct ll_d_hlist_node *tmp;
646 int dentry_count = 0;
648 LASSERT(inode != NULL);
650 ll_d_hlist_for_each(tmp, &inode->i_dentry)
653 CERROR("inode %p dump: dev=%s ino=%lu mode=%o count=%u, %d dentries\n",
654 inode, ll_i2mdexp(inode)->exp_obd->obd_name, inode->i_ino,
655 inode->i_mode, atomic_read(&inode->i_count), dentry_count);
658 void lustre_dump_dentry(struct dentry *dentry, int recur)
660 struct list_head *tmp;
663 LASSERT(dentry != NULL);
665 list_for_each(tmp, &dentry->d_subdirs)
668 CERROR("dentry %p dump: name=%.*s parent=%.*s (%p), inode=%p, count=%u,"
669 " flags=0x%x, fsdata=%p, %d subdirs\n", dentry,
670 dentry->d_name.len, dentry->d_name.name,
671 dentry->d_parent->d_name.len, dentry->d_parent->d_name.name,
672 dentry->d_parent, dentry->d_inode, d_count(dentry),
673 dentry->d_flags, dentry->d_fsdata, subdirs);
674 if (dentry->d_inode != NULL)
675 ll_dump_inode(dentry->d_inode);
680 list_for_each(tmp, &dentry->d_subdirs) {
681 struct dentry *d = list_entry(tmp, struct dentry, d_u.d_child);
682 lustre_dump_dentry(d, recur - 1);
686 void client_common_put_super(struct super_block *sb)
688 struct ll_sb_info *sbi = ll_s2sbi(sb);
690 #ifdef CONFIG_FS_POSIX_ACL
691 if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
692 et_fini(&sbi->ll_et);
693 rct_fini(&sbi->ll_rct);
697 ll_close_thread_shutdown(sbi->ll_lcq);
701 list_del(&sbi->ll_conn_chain);
703 obd_fid_fini(sbi->ll_dt_exp->exp_obd);
704 obd_disconnect(sbi->ll_dt_exp);
705 sbi->ll_dt_exp = NULL;
706 /* wait till all OSCs are gone, since cl_cache is accessing sbi.
708 obd_zombie_barrier();
710 lprocfs_unregister_mountpoint(sbi);
712 obd_fid_fini(sbi->ll_md_exp->exp_obd);
713 obd_disconnect(sbi->ll_md_exp);
714 sbi->ll_md_exp = NULL;
717 void ll_kill_super(struct super_block *sb)
719 struct ll_sb_info *sbi;
722 if (!(sb->s_flags & MS_ACTIVE))
726 /* we need restore s_dev from changed for clustred NFS before put_super
727 * because new kernels have cached s_dev and change sb->s_dev in
728 * put_super not affected real removing devices */
730 sb->s_dev = sbi->ll_sdev_orig;
731 sbi->ll_umounting = 1;
735 char *ll_read_opt(const char *opt, char *data)
740 CDEBUG(D_SUPER, "option: %s, data %s\n", opt, data);
741 if (strncmp(opt, data, strlen(opt)))
743 if ((value = strchr(data, '=')) == NULL)
747 OBD_ALLOC(retval, strlen(value) + 1);
749 CERROR("out of memory!\n");
753 memcpy(retval, value, strlen(value)+1);
754 CDEBUG(D_SUPER, "Assigned option: %s, value %s\n", opt, retval);
758 static inline int ll_set_opt(const char *opt, char *data, int fl)
760 if (strncmp(opt, data, strlen(opt)) != 0)
766 /* non-client-specific mount options are parsed in lmd_parse */
767 static int ll_options(char *options, int *flags)
770 char *s1 = options, *s2;
775 CDEBUG(D_CONFIG, "Parsing opts %s\n", options);
778 CDEBUG(D_SUPER, "next opt=%s\n", s1);
779 tmp = ll_set_opt("nolock", s1, LL_SBI_NOLCK);
784 tmp = ll_set_opt("flock", s1, LL_SBI_FLOCK);
789 tmp = ll_set_opt("localflock", s1, LL_SBI_LOCALFLOCK);
794 tmp = ll_set_opt("noflock", s1, LL_SBI_FLOCK|LL_SBI_LOCALFLOCK);
799 tmp = ll_set_opt("user_xattr", s1, LL_SBI_USER_XATTR);
804 tmp = ll_set_opt("nouser_xattr", s1, LL_SBI_USER_XATTR);
809 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 5, 50, 0)
810 tmp = ll_set_opt("acl", s1, LL_SBI_ACL);
812 /* Ignore deprecated mount option. The client will
813 * always try to mount with ACL support, whether this
814 * is used depends on whether server supports it. */
815 LCONSOLE_ERROR_MSG(0x152, "Ignoring deprecated "
816 "mount option 'acl'.\n");
819 tmp = ll_set_opt("noacl", s1, LL_SBI_ACL);
821 LCONSOLE_ERROR_MSG(0x152, "Ignoring deprecated "
822 "mount option 'noacl'.\n");
826 #warning "{no}acl options have been deprecated since 1.8, please remove them"
828 tmp = ll_set_opt("remote_client", s1, LL_SBI_RMT_CLIENT);
833 tmp = ll_set_opt("user_fid2path", s1, LL_SBI_USER_FID2PATH);
838 tmp = ll_set_opt("nouser_fid2path", s1, LL_SBI_USER_FID2PATH);
844 tmp = ll_set_opt("checksum", s1, LL_SBI_CHECKSUM);
849 tmp = ll_set_opt("nochecksum", s1, LL_SBI_CHECKSUM);
854 tmp = ll_set_opt("lruresize", s1, LL_SBI_LRU_RESIZE);
859 tmp = ll_set_opt("nolruresize", s1, LL_SBI_LRU_RESIZE);
864 tmp = ll_set_opt("lazystatfs", s1, LL_SBI_LAZYSTATFS);
869 tmp = ll_set_opt("nolazystatfs", s1, LL_SBI_LAZYSTATFS);
874 tmp = ll_set_opt("som_preview", s1, LL_SBI_SOM_PREVIEW);
879 tmp = ll_set_opt("32bitapi", s1, LL_SBI_32BIT_API);
884 tmp = ll_set_opt("verbose", s1, LL_SBI_VERBOSE);
889 tmp = ll_set_opt("noverbose", s1, LL_SBI_VERBOSE);
894 LCONSOLE_ERROR_MSG(0x152, "Unknown option '%s', won't mount.\n",
900 s2 = strchr(s1, ',');
908 void ll_lli_init(struct ll_inode_info *lli)
910 lli->lli_inode_magic = LLI_INODE_MAGIC;
912 lli->lli_ioepoch = 0;
913 lli->lli_maxbytes = MAX_LFS_FILESIZE;
914 spin_lock_init(&lli->lli_lock);
915 lli->lli_posix_acl = NULL;
916 lli->lli_remote_perms = NULL;
917 mutex_init(&lli->lli_rmtperm_mutex);
918 /* Do not set lli_fid, it has been initialized already. */
919 fid_zero(&lli->lli_pfid);
920 INIT_LIST_HEAD(&lli->lli_close_list);
921 INIT_LIST_HEAD(&lli->lli_oss_capas);
922 atomic_set(&lli->lli_open_count, 0);
923 lli->lli_mds_capa = NULL;
924 lli->lli_rmtperm_time = 0;
925 lli->lli_pending_och = NULL;
926 lli->lli_mds_read_och = NULL;
927 lli->lli_mds_write_och = NULL;
928 lli->lli_mds_exec_och = NULL;
929 lli->lli_open_fd_read_count = 0;
930 lli->lli_open_fd_write_count = 0;
931 lli->lli_open_fd_exec_count = 0;
932 mutex_init(&lli->lli_och_mutex);
933 spin_lock_init(&lli->lli_agl_lock);
934 lli->lli_has_smd = false;
935 lli->lli_layout_gen = LL_LAYOUT_GEN_NONE;
936 lli->lli_clob = NULL;
938 init_rwsem(&lli->lli_xattrs_list_rwsem);
939 mutex_init(&lli->lli_xattrs_enq_lock);
941 LASSERT(lli->lli_vfs_inode.i_mode != 0);
942 if (S_ISDIR(lli->lli_vfs_inode.i_mode)) {
943 mutex_init(&lli->lli_readdir_mutex);
944 lli->lli_opendir_key = NULL;
946 lli->lli_def_acl = NULL;
947 spin_lock_init(&lli->lli_sa_lock);
948 lli->lli_opendir_pid = 0;
950 sema_init(&lli->lli_size_sem, 1);
951 lli->lli_size_sem_owner = NULL;
952 lli->lli_symlink_name = NULL;
953 init_rwsem(&lli->lli_trunc_sem);
954 mutex_init(&lli->lli_write_mutex);
955 init_rwsem(&lli->lli_glimpse_sem);
956 lli->lli_glimpse_time = 0;
957 INIT_LIST_HEAD(&lli->lli_agl_list);
958 lli->lli_agl_index = 0;
959 lli->lli_async_rc = 0;
960 lli->lli_volatile = false;
962 mutex_init(&lli->lli_layout_mutex);
965 static inline int ll_bdi_register(struct backing_dev_info *bdi)
967 static atomic_t ll_bdi_num = ATOMIC_INIT(0);
969 bdi->name = "lustre";
970 return bdi_register(bdi, NULL, "lustre-%d",
971 atomic_inc_return(&ll_bdi_num));
974 int ll_fill_super(struct super_block *sb, struct vfsmount *mnt)
976 struct lustre_profile *lprof = NULL;
977 struct lustre_sb_info *lsi = s2lsi(sb);
978 struct ll_sb_info *sbi;
979 char *dt = NULL, *md = NULL;
980 char *profilenm = get_profile_name(sb);
981 struct config_llog_instance *cfg;
982 /* %p for void* in printf needs 16+2 characters: 0xffffffffffffffff */
983 const int instlen = sizeof(cfg->cfg_instance) * 2 + 2;
986 CDEBUG(D_VFSTRACE, "VFS Op: sb %p\n", sb);
992 try_module_get(THIS_MODULE);
994 /* client additional sb info */
995 lsi->lsi_llsbi = sbi = ll_init_sbi();
997 module_put(THIS_MODULE);
1002 err = ll_options(lsi->lsi_lmd->lmd_opts, &sbi->ll_flags);
1004 GOTO(out_free, err);
1006 err = bdi_init(&lsi->lsi_bdi);
1008 GOTO(out_free, err);
1009 lsi->lsi_flags |= LSI_BDI_INITIALIZED;
1010 lsi->lsi_bdi.capabilities = BDI_CAP_MAP_COPY;
1011 err = ll_bdi_register(&lsi->lsi_bdi);
1013 GOTO(out_free, err);
1015 sb->s_bdi = &lsi->lsi_bdi;
1017 /* Generate a string unique to this super, in case some joker tries
1018 to mount the same fs at two mount points.
1019 Use the address of the super itself.*/
1020 cfg->cfg_instance = sb;
1021 cfg->cfg_uuid = lsi->lsi_llsbi->ll_sb_uuid;
1022 cfg->cfg_callback = class_config_llog_handler;
1023 /* set up client obds */
1024 err = lustre_process_log(sb, profilenm, cfg);
1026 CERROR("Unable to process log: %d\n", err);
1027 GOTO(out_free, err);
1030 /* Profile set with LCFG_MOUNTOPT so we can find our mdc and osc obds */
1031 lprof = class_get_profile(profilenm);
1032 if (lprof == NULL) {
1033 LCONSOLE_ERROR_MSG(0x156, "The client profile '%s' could not be"
1034 " read from the MGS. Does that filesystem "
1035 "exist?\n", profilenm);
1036 GOTO(out_free, err = -EINVAL);
1038 CDEBUG(D_CONFIG, "Found profile %s: mdc=%s osc=%s\n", profilenm,
1039 lprof->lp_md, lprof->lp_dt);
1041 OBD_ALLOC(dt, strlen(lprof->lp_dt) + instlen + 2);
1043 GOTO(out_free, err = -ENOMEM);
1044 sprintf(dt, "%s-%p", lprof->lp_dt, cfg->cfg_instance);
1046 OBD_ALLOC(md, strlen(lprof->lp_md) + instlen + 2);
1048 GOTO(out_free, err = -ENOMEM);
1049 sprintf(md, "%s-%p", lprof->lp_md, cfg->cfg_instance);
1051 /* connections, registrations, sb setup */
1052 err = client_common_fill_super(sb, md, dt, mnt);
1056 OBD_FREE(md, strlen(lprof->lp_md) + instlen + 2);
1058 OBD_FREE(dt, strlen(lprof->lp_dt) + instlen + 2);
1061 else if (sbi->ll_flags & LL_SBI_VERBOSE)
1062 LCONSOLE_WARN("Mounted %s\n", profilenm);
1066 } /* ll_fill_super */
1068 void ll_put_super(struct super_block *sb)
1070 struct config_llog_instance cfg;
1071 struct obd_device *obd;
1072 struct lustre_sb_info *lsi = s2lsi(sb);
1073 struct ll_sb_info *sbi = ll_s2sbi(sb);
1074 char *profilenm = get_profile_name(sb);
1075 int next, force = 1;
1077 CDEBUG(D_VFSTRACE, "VFS Op: sb %p - %s\n", sb, profilenm);
1079 ll_print_capa_stat(sbi);
1081 cfg.cfg_instance = sb;
1082 lustre_end_log(sb, profilenm, &cfg);
1084 if (sbi->ll_md_exp) {
1085 obd = class_exp2obd(sbi->ll_md_exp);
1087 force = obd->obd_force;
1090 /* We need to set force before the lov_disconnect in
1091 lustre_common_put_super, since l_d cleans up osc's as well. */
1094 while ((obd = class_devices_in_group(&sbi->ll_sb_uuid,
1096 obd->obd_force = force;
1101 /* Only if client_common_fill_super succeeded */
1102 client_common_put_super(sb);
1106 while ((obd = class_devices_in_group(&sbi->ll_sb_uuid, &next)) !=NULL) {
1107 class_manual_cleanup(obd);
1110 if (sbi->ll_flags & LL_SBI_VERBOSE)
1111 LCONSOLE_WARN("Unmounted %s\n", profilenm ? profilenm : "");
1114 class_del_profile(profilenm);
1116 if (lsi->lsi_flags & LSI_BDI_INITIALIZED) {
1117 bdi_destroy(&lsi->lsi_bdi);
1118 lsi->lsi_flags &= ~LSI_BDI_INITIALIZED;
1122 lsi->lsi_llsbi = NULL;
1124 lustre_common_put_super(sb);
1126 module_put(THIS_MODULE);
1127 } /* client_put_super */
1129 struct inode *ll_inode_from_resource_lock(struct ldlm_lock *lock)
1131 struct inode *inode = NULL;
1133 /* NOTE: we depend on atomic igrab() -bzzz */
1134 lock_res_and_lock(lock);
1135 if (lock->l_resource->lr_lvb_inode) {
1136 struct ll_inode_info * lli;
1137 lli = ll_i2info(lock->l_resource->lr_lvb_inode);
1138 if (lli->lli_inode_magic == LLI_INODE_MAGIC) {
1139 inode = igrab(lock->l_resource->lr_lvb_inode);
1141 inode = lock->l_resource->lr_lvb_inode;
1142 LDLM_DEBUG_LIMIT(inode->i_state & I_FREEING ? D_INFO :
1143 D_WARNING, lock, "lr_lvb_inode %p is "
1144 "bogus: magic %08x",
1145 lock->l_resource->lr_lvb_inode,
1146 lli->lli_inode_magic);
1150 unlock_res_and_lock(lock);
1154 struct inode *ll_inode_from_lock(struct ldlm_lock *lock)
1156 struct inode *inode = NULL;
1157 /* NOTE: we depend on atomic igrab() -bzzz */
1158 lock_res_and_lock(lock);
1159 if (lock->l_ast_data) {
1160 struct ll_inode_info *lli = ll_i2info(lock->l_ast_data);
1161 if (lli->lli_inode_magic == LLI_INODE_MAGIC) {
1162 inode = igrab(lock->l_ast_data);
1164 inode = lock->l_ast_data;
1165 LDLM_DEBUG_LIMIT(inode->i_state & I_FREEING ? D_INFO :
1166 D_WARNING, lock, "l_ast_data %p is "
1167 "bogus: magic %08x", lock->l_ast_data,
1168 lli->lli_inode_magic);
1172 unlock_res_and_lock(lock);
1176 void ll_clear_inode(struct inode *inode)
1178 struct ll_inode_info *lli = ll_i2info(inode);
1179 struct ll_sb_info *sbi = ll_i2sbi(inode);
1181 CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p)\n", inode->i_ino,
1182 inode->i_generation, inode);
1184 if (S_ISDIR(inode->i_mode)) {
1185 /* these should have been cleared in ll_file_release */
1186 LASSERT(lli->lli_opendir_key == NULL);
1187 LASSERT(lli->lli_sai == NULL);
1188 LASSERT(lli->lli_opendir_pid == 0);
1191 spin_lock(&lli->lli_lock);
1192 ll_i2info(inode)->lli_flags &= ~LLIF_MDS_SIZE_LOCK;
1193 spin_unlock(&lli->lli_lock);
1194 md_null_inode(sbi->ll_md_exp, ll_inode2fid(inode));
1196 LASSERT(!lli->lli_open_fd_write_count);
1197 LASSERT(!lli->lli_open_fd_read_count);
1198 LASSERT(!lli->lli_open_fd_exec_count);
1200 if (lli->lli_mds_write_och)
1201 ll_md_real_close(inode, FMODE_WRITE);
1202 if (lli->lli_mds_exec_och)
1203 ll_md_real_close(inode, FMODE_EXEC);
1204 if (lli->lli_mds_read_och)
1205 ll_md_real_close(inode, FMODE_READ);
1207 if (S_ISLNK(inode->i_mode) && lli->lli_symlink_name) {
1208 OBD_FREE(lli->lli_symlink_name,
1209 strlen(lli->lli_symlink_name) + 1);
1210 lli->lli_symlink_name = NULL;
1213 ll_xattr_cache_destroy(inode);
1215 if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
1216 LASSERT(lli->lli_posix_acl == NULL);
1217 if (lli->lli_remote_perms) {
1218 free_rmtperm_hash(lli->lli_remote_perms);
1219 lli->lli_remote_perms = NULL;
1222 #ifdef CONFIG_FS_POSIX_ACL
1223 else if (lli->lli_posix_acl) {
1224 LASSERT(atomic_read(&lli->lli_posix_acl->a_refcount) == 1);
1225 LASSERT(lli->lli_remote_perms == NULL);
1226 posix_acl_release(lli->lli_posix_acl);
1227 lli->lli_posix_acl = NULL;
1230 lli->lli_inode_magic = LLI_INODE_DEAD;
1232 ll_clear_inode_capas(inode);
1233 if (!S_ISDIR(inode->i_mode))
1234 LASSERT(list_empty(&lli->lli_agl_list));
1237 * XXX This has to be done before lsm is freed below, because
1238 * cl_object still uses inode lsm.
1240 cl_inode_fini(inode);
1241 lli->lli_has_smd = false;
1244 int ll_md_setattr(struct dentry *dentry, struct md_op_data *op_data,
1245 struct md_open_data **mod)
1247 struct lustre_md md;
1248 struct inode *inode = dentry->d_inode;
1249 struct ll_sb_info *sbi = ll_i2sbi(inode);
1250 struct ptlrpc_request *request = NULL;
1253 op_data = ll_prep_md_op_data(op_data, inode, NULL, NULL, 0, 0,
1254 LUSTRE_OPC_ANY, NULL);
1255 if (IS_ERR(op_data))
1256 return PTR_ERR(op_data);
1258 rc = md_setattr(sbi->ll_md_exp, op_data, NULL, 0, NULL, 0,
1261 ptlrpc_req_finished(request);
1262 if (rc == -ENOENT) {
1264 /* Unlinked special device node? Or just a race?
1265 * Pretend we done everything. */
1266 if (!S_ISREG(inode->i_mode) &&
1267 !S_ISDIR(inode->i_mode)) {
1268 ia_valid = op_data->op_attr.ia_valid;
1269 op_data->op_attr.ia_valid &= ~TIMES_SET_FLAGS;
1270 rc = simple_setattr(dentry, &op_data->op_attr);
1271 op_data->op_attr.ia_valid = ia_valid;
1273 } else if (rc != -EPERM && rc != -EACCES && rc != -ETXTBSY) {
1274 CERROR("md_setattr fails: rc = %d\n", rc);
1279 rc = md_get_lustre_md(sbi->ll_md_exp, request, sbi->ll_dt_exp,
1280 sbi->ll_md_exp, &md);
1282 ptlrpc_req_finished(request);
1286 ia_valid = op_data->op_attr.ia_valid;
1287 /* inode size will be in ll_setattr_ost, can't do it now since dirty
1288 * cache is not cleared yet. */
1289 op_data->op_attr.ia_valid &= ~(TIMES_SET_FLAGS | ATTR_SIZE);
1290 rc = simple_setattr(dentry, &op_data->op_attr);
1291 op_data->op_attr.ia_valid = ia_valid;
1293 /* Extract epoch data if obtained. */
1294 op_data->op_handle = md.body->handle;
1295 op_data->op_ioepoch = md.body->ioepoch;
1297 ll_update_inode(inode, &md);
1298 ptlrpc_req_finished(request);
1303 /* Close IO epoch and send Size-on-MDS attribute update. */
1304 static int ll_setattr_done_writing(struct inode *inode,
1305 struct md_op_data *op_data,
1306 struct md_open_data *mod)
1308 struct ll_inode_info *lli = ll_i2info(inode);
1311 LASSERT(op_data != NULL);
1312 if (!S_ISREG(inode->i_mode))
1315 CDEBUG(D_INODE, "Epoch "LPU64" closed on "DFID" for truncate\n",
1316 op_data->op_ioepoch, PFID(&lli->lli_fid));
1318 op_data->op_flags = MF_EPOCH_CLOSE;
1319 ll_done_writing_attr(inode, op_data);
1320 ll_pack_inode2opdata(inode, op_data, NULL);
1322 rc = md_done_writing(ll_i2sbi(inode)->ll_md_exp, op_data, mod);
1323 if (rc == -EAGAIN) {
1324 /* MDS has instructed us to obtain Size-on-MDS attribute
1325 * from OSTs and send setattr to back to MDS. */
1326 rc = ll_som_update(inode, op_data);
1328 CERROR("inode %lu mdc truncate failed: rc = %d\n",
1334 static int ll_setattr_ost(struct inode *inode, struct iattr *attr)
1336 struct obd_capa *capa;
1339 if (attr->ia_valid & ATTR_SIZE)
1340 capa = ll_osscapa_get(inode, CAPA_OPC_OSS_TRUNC);
1342 capa = ll_mdscapa_get(inode);
1344 rc = cl_setattr_ost(inode, attr, capa);
1346 if (attr->ia_valid & ATTR_SIZE)
1347 ll_truncate_free_capa(capa);
1355 /* If this inode has objects allocated to it (lsm != NULL), then the OST
1356 * object(s) determine the file size and mtime. Otherwise, the MDS will
1357 * keep these values until such a time that objects are allocated for it.
1358 * We do the MDS operations first, as it is checking permissions for us.
1359 * We don't to the MDS RPC if there is nothing that we want to store there,
1360 * otherwise there is no harm in updating mtime/atime on the MDS if we are
1361 * going to do an RPC anyways.
1363 * If we are doing a truncate, we will send the mtime and ctime updates
1364 * to the OST with the punch RPC, otherwise we do an explicit setattr RPC.
1365 * I don't believe it is possible to get e.g. ATTR_MTIME_SET and ATTR_SIZE
1368 * In case of HSMimport, we only set attr on MDS.
1370 int ll_setattr_raw(struct dentry *dentry, struct iattr *attr, bool hsm_import)
1372 struct inode *inode = dentry->d_inode;
1373 struct ll_inode_info *lli = ll_i2info(inode);
1374 struct md_op_data *op_data = NULL;
1375 struct md_open_data *mod = NULL;
1376 bool file_is_released = false;
1377 int rc = 0, rc1 = 0;
1380 "%s: setattr inode %p/fid:"DFID
1381 " from %llu to %llu, valid %x, hsm_import %d\n",
1382 ll_get_fsname(inode->i_sb, NULL, 0), inode,
1383 PFID(&lli->lli_fid), i_size_read(inode), attr->ia_size,
1384 attr->ia_valid, hsm_import);
1386 if (attr->ia_valid & ATTR_SIZE) {
1387 /* Check new size against VFS/VM file size limit and rlimit */
1388 rc = inode_newsize_ok(inode, attr->ia_size);
1392 /* The maximum Lustre file size is variable, based on the
1393 * OST maximum object size and number of stripes. This
1394 * needs another check in addition to the VFS check above. */
1395 if (attr->ia_size > ll_file_maxbytes(inode)) {
1396 CDEBUG(D_INODE,"file "DFID" too large %llu > "LPU64"\n",
1397 PFID(&lli->lli_fid), attr->ia_size,
1398 ll_file_maxbytes(inode));
1402 attr->ia_valid |= ATTR_MTIME | ATTR_CTIME;
1405 /* POSIX: check before ATTR_*TIME_SET set (from inode_change_ok) */
1406 if (attr->ia_valid & TIMES_SET_FLAGS) {
1407 if ((!uid_eq(current_fsuid(), inode->i_uid)) &&
1408 !cfs_capable(CFS_CAP_FOWNER))
1412 /* We mark all of the fields "set" so MDS/OST does not re-set them */
1413 if (attr->ia_valid & ATTR_CTIME) {
1414 attr->ia_ctime = CURRENT_TIME;
1415 attr->ia_valid |= ATTR_CTIME_SET;
1417 if (!(attr->ia_valid & ATTR_ATIME_SET) &&
1418 (attr->ia_valid & ATTR_ATIME)) {
1419 attr->ia_atime = CURRENT_TIME;
1420 attr->ia_valid |= ATTR_ATIME_SET;
1422 if (!(attr->ia_valid & ATTR_MTIME_SET) &&
1423 (attr->ia_valid & ATTR_MTIME)) {
1424 attr->ia_mtime = CURRENT_TIME;
1425 attr->ia_valid |= ATTR_MTIME_SET;
1428 if (attr->ia_valid & (ATTR_MTIME | ATTR_CTIME))
1429 CDEBUG(D_INODE, "setting mtime %lu, ctime %lu, now = %lu\n",
1430 LTIME_S(attr->ia_mtime), LTIME_S(attr->ia_ctime),
1431 cfs_time_current_sec());
1433 /* If we are changing file size, file content is modified, flag it. */
1434 if (attr->ia_valid & ATTR_SIZE) {
1435 attr->ia_valid |= MDS_OPEN_OWNEROVERRIDE;
1436 spin_lock(&lli->lli_lock);
1437 lli->lli_flags |= LLIF_DATA_MODIFIED;
1438 spin_unlock(&lli->lli_lock);
1441 /* We always do an MDS RPC, even if we're only changing the size;
1442 * only the MDS knows whether truncate() should fail with -ETXTBUSY */
1444 OBD_ALLOC_PTR(op_data);
1445 if (op_data == NULL)
1448 if (!S_ISDIR(inode->i_mode)) {
1449 if (attr->ia_valid & ATTR_SIZE)
1450 inode_dio_write_done(inode);
1451 mutex_unlock(&inode->i_mutex);
1452 down_write(&lli->lli_trunc_sem);
1455 memcpy(&op_data->op_attr, attr, sizeof(*attr));
1457 /* Open epoch for truncate. */
1458 if (exp_connect_som(ll_i2mdexp(inode)) &&
1459 (attr->ia_valid & (ATTR_SIZE | ATTR_MTIME | ATTR_MTIME_SET)))
1460 op_data->op_flags = MF_EPOCH_OPEN;
1462 /* truncate on a released file must failed with -ENODATA,
1463 * so size must not be set on MDS for released file
1464 * but other attributes must be set
1466 if (S_ISREG(inode->i_mode)) {
1467 struct lov_stripe_md *lsm;
1470 ll_layout_refresh(inode, &gen);
1471 lsm = ccc_inode_lsm_get(inode);
1472 if (lsm && lsm->lsm_pattern & LOV_PATTERN_F_RELEASED)
1473 file_is_released = true;
1474 ccc_inode_lsm_put(inode, lsm);
1477 /* if not in HSM import mode, clear size attr for released file
1478 * we clear the attribute send to MDT in op_data, not the original
1479 * received from caller in attr which is used later to
1480 * decide return code */
1481 if (file_is_released && (attr->ia_valid & ATTR_SIZE) && !hsm_import)
1482 op_data->op_attr.ia_valid &= ~ATTR_SIZE;
1484 rc = ll_md_setattr(dentry, op_data, &mod);
1488 /* truncate failed (only when non HSM import), others succeed */
1489 if (file_is_released) {
1490 if ((attr->ia_valid & ATTR_SIZE) && !hsm_import)
1491 GOTO(out, rc = -ENODATA);
1496 /* RPC to MDT is sent, cancel data modification flag */
1497 if (rc == 0 && (op_data->op_bias & MDS_DATA_MODIFIED)) {
1498 spin_lock(&lli->lli_lock);
1499 lli->lli_flags &= ~LLIF_DATA_MODIFIED;
1500 spin_unlock(&lli->lli_lock);
1503 ll_ioepoch_open(lli, op_data->op_ioepoch);
1504 if (!S_ISREG(inode->i_mode))
1507 if (attr->ia_valid & (ATTR_SIZE |
1508 ATTR_ATIME | ATTR_ATIME_SET |
1509 ATTR_MTIME | ATTR_MTIME_SET))
1510 /* For truncate and utimes sending attributes to OSTs, setting
1511 * mtime/atime to the past will be performed under PW [0:EOF]
1512 * extent lock (new_size:EOF for truncate). It may seem
1513 * excessive to send mtime/atime updates to OSTs when not
1514 * setting times to past, but it is necessary due to possible
1515 * time de-synchronization between MDT inode and OST objects */
1516 rc = ll_setattr_ost(inode, attr);
1519 if (op_data->op_ioepoch) {
1520 rc1 = ll_setattr_done_writing(inode, op_data, mod);
1524 ll_finish_md_op_data(op_data);
1526 if (!S_ISDIR(inode->i_mode)) {
1527 up_write(&lli->lli_trunc_sem);
1528 mutex_lock(&inode->i_mutex);
1529 if ((attr->ia_valid & ATTR_SIZE) && !hsm_import)
1530 inode_dio_wait(inode);
1533 ll_stats_ops_tally(ll_i2sbi(inode), (attr->ia_valid & ATTR_SIZE) ?
1534 LPROC_LL_TRUNC : LPROC_LL_SETATTR, 1);
1539 int ll_setattr(struct dentry *de, struct iattr *attr)
1541 int mode = de->d_inode->i_mode;
1543 if ((attr->ia_valid & (ATTR_CTIME|ATTR_SIZE|ATTR_MODE)) ==
1544 (ATTR_CTIME|ATTR_SIZE|ATTR_MODE))
1545 attr->ia_valid |= MDS_OPEN_OWNEROVERRIDE;
1547 if (((attr->ia_valid & (ATTR_MODE|ATTR_FORCE|ATTR_SIZE)) ==
1548 (ATTR_SIZE|ATTR_MODE)) &&
1549 (((mode & S_ISUID) && !(attr->ia_mode & S_ISUID)) ||
1550 (((mode & (S_ISGID|S_IXGRP)) == (S_ISGID|S_IXGRP)) &&
1551 !(attr->ia_mode & S_ISGID))))
1552 attr->ia_valid |= ATTR_FORCE;
1554 if ((mode & S_ISUID) &&
1555 !(attr->ia_mode & S_ISUID) &&
1556 !(attr->ia_valid & ATTR_KILL_SUID))
1557 attr->ia_valid |= ATTR_KILL_SUID;
1559 if (((mode & (S_ISGID|S_IXGRP)) == (S_ISGID|S_IXGRP)) &&
1560 !(attr->ia_mode & S_ISGID) &&
1561 !(attr->ia_valid & ATTR_KILL_SGID))
1562 attr->ia_valid |= ATTR_KILL_SGID;
1564 return ll_setattr_raw(de, attr, false);
1567 int ll_statfs_internal(struct super_block *sb, struct obd_statfs *osfs,
1568 __u64 max_age, __u32 flags)
1570 struct ll_sb_info *sbi = ll_s2sbi(sb);
1571 struct obd_statfs obd_osfs;
1574 rc = obd_statfs(NULL, sbi->ll_md_exp, osfs, max_age, flags);
1576 CERROR("md_statfs fails: rc = %d\n", rc);
1580 osfs->os_type = sb->s_magic;
1582 CDEBUG(D_SUPER, "MDC blocks "LPU64"/"LPU64" objects "LPU64"/"LPU64"\n",
1583 osfs->os_bavail, osfs->os_blocks, osfs->os_ffree,osfs->os_files);
1585 if (sbi->ll_flags & LL_SBI_LAZYSTATFS)
1586 flags |= OBD_STATFS_NODELAY;
1588 rc = obd_statfs_rqset(sbi->ll_dt_exp, &obd_osfs, max_age, flags);
1590 CERROR("obd_statfs fails: rc = %d\n", rc);
1594 CDEBUG(D_SUPER, "OSC blocks "LPU64"/"LPU64" objects "LPU64"/"LPU64"\n",
1595 obd_osfs.os_bavail, obd_osfs.os_blocks, obd_osfs.os_ffree,
1598 osfs->os_bsize = obd_osfs.os_bsize;
1599 osfs->os_blocks = obd_osfs.os_blocks;
1600 osfs->os_bfree = obd_osfs.os_bfree;
1601 osfs->os_bavail = obd_osfs.os_bavail;
1603 /* If we don't have as many objects free on the OST as inodes
1604 * on the MDS, we reduce the total number of inodes to
1605 * compensate, so that the "inodes in use" number is correct.
1607 if (obd_osfs.os_ffree < osfs->os_ffree) {
1608 osfs->os_files = (osfs->os_files - osfs->os_ffree) +
1610 osfs->os_ffree = obd_osfs.os_ffree;
1615 int ll_statfs(struct dentry *de, struct kstatfs *sfs)
1617 struct super_block *sb = de->d_sb;
1618 struct obd_statfs osfs;
1621 CDEBUG(D_VFSTRACE, "VFS Op: at "LPU64" jiffies\n", get_jiffies_64());
1622 ll_stats_ops_tally(ll_s2sbi(sb), LPROC_LL_STAFS, 1);
1624 /* Some amount of caching on the client is allowed */
1625 rc = ll_statfs_internal(sb, &osfs,
1626 cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS),
1631 statfs_unpack(sfs, &osfs);
1633 /* We need to downshift for all 32-bit kernels, because we can't
1634 * tell if the kernel is being called via sys_statfs64() or not.
1635 * Stop before overflowing f_bsize - in which case it is better
1636 * to just risk EOVERFLOW if caller is using old sys_statfs(). */
1637 if (sizeof(long) < 8) {
1638 while (osfs.os_blocks > ~0UL && sfs->f_bsize < 0x40000000) {
1641 osfs.os_blocks >>= 1;
1642 osfs.os_bfree >>= 1;
1643 osfs.os_bavail >>= 1;
1647 sfs->f_blocks = osfs.os_blocks;
1648 sfs->f_bfree = osfs.os_bfree;
1649 sfs->f_bavail = osfs.os_bavail;
1650 sfs->f_fsid = ll_s2sbi(sb)->ll_fsid;
1654 void ll_inode_size_lock(struct inode *inode)
1656 struct ll_inode_info *lli;
1658 LASSERT(!S_ISDIR(inode->i_mode));
1660 lli = ll_i2info(inode);
1661 LASSERT(lli->lli_size_sem_owner != current);
1662 down(&lli->lli_size_sem);
1663 LASSERT(lli->lli_size_sem_owner == NULL);
1664 lli->lli_size_sem_owner = current;
1667 void ll_inode_size_unlock(struct inode *inode)
1669 struct ll_inode_info *lli;
1671 lli = ll_i2info(inode);
1672 LASSERT(lli->lli_size_sem_owner == current);
1673 lli->lli_size_sem_owner = NULL;
1674 up(&lli->lli_size_sem);
1677 void ll_update_inode(struct inode *inode, struct lustre_md *md)
1679 struct ll_inode_info *lli = ll_i2info(inode);
1680 struct mdt_body *body = md->body;
1681 struct lov_stripe_md *lsm = md->lsm;
1682 struct ll_sb_info *sbi = ll_i2sbi(inode);
1684 LASSERT ((lsm != NULL) == ((body->valid & OBD_MD_FLEASIZE) != 0));
1686 if (!lli->lli_has_smd &&
1687 !(sbi->ll_flags & LL_SBI_LAYOUT_LOCK))
1688 cl_file_inode_init(inode, md);
1690 lli->lli_maxbytes = lsm->lsm_maxbytes;
1691 if (lli->lli_maxbytes > MAX_LFS_FILESIZE)
1692 lli->lli_maxbytes = MAX_LFS_FILESIZE;
1695 if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
1696 if (body->valid & OBD_MD_FLRMTPERM)
1697 ll_update_remote_perm(inode, md->remote_perm);
1699 #ifdef CONFIG_FS_POSIX_ACL
1700 else if (body->valid & OBD_MD_FLACL) {
1701 spin_lock(&lli->lli_lock);
1702 if (lli->lli_posix_acl)
1703 posix_acl_release(lli->lli_posix_acl);
1704 lli->lli_posix_acl = md->posix_acl;
1705 spin_unlock(&lli->lli_lock);
1708 inode->i_ino = cl_fid_build_ino(&body->fid1,
1709 sbi->ll_flags & LL_SBI_32BIT_API);
1710 inode->i_generation = cl_fid_build_gen(&body->fid1);
1712 if (body->valid & OBD_MD_FLATIME) {
1713 if (body->atime > LTIME_S(inode->i_atime))
1714 LTIME_S(inode->i_atime) = body->atime;
1715 lli->lli_lvb.lvb_atime = body->atime;
1717 if (body->valid & OBD_MD_FLMTIME) {
1718 if (body->mtime > LTIME_S(inode->i_mtime)) {
1719 CDEBUG(D_INODE, "setting ino %lu mtime from %lu "
1720 "to "LPU64"\n", inode->i_ino,
1721 LTIME_S(inode->i_mtime), body->mtime);
1722 LTIME_S(inode->i_mtime) = body->mtime;
1724 lli->lli_lvb.lvb_mtime = body->mtime;
1726 if (body->valid & OBD_MD_FLCTIME) {
1727 if (body->ctime > LTIME_S(inode->i_ctime))
1728 LTIME_S(inode->i_ctime) = body->ctime;
1729 lli->lli_lvb.lvb_ctime = body->ctime;
1731 if (body->valid & OBD_MD_FLMODE)
1732 inode->i_mode = (inode->i_mode & S_IFMT)|(body->mode & ~S_IFMT);
1733 if (body->valid & OBD_MD_FLTYPE)
1734 inode->i_mode = (inode->i_mode & ~S_IFMT)|(body->mode & S_IFMT);
1735 LASSERT(inode->i_mode != 0);
1736 if (S_ISREG(inode->i_mode)) {
1737 inode->i_blkbits = min(PTLRPC_MAX_BRW_BITS + 1, LL_MAX_BLKSIZE_BITS);
1739 inode->i_blkbits = inode->i_sb->s_blocksize_bits;
1741 if (body->valid & OBD_MD_FLUID)
1742 inode->i_uid = make_kuid(&init_user_ns, body->uid);
1743 if (body->valid & OBD_MD_FLGID)
1744 inode->i_gid = make_kgid(&init_user_ns, body->gid);
1745 if (body->valid & OBD_MD_FLFLAGS)
1746 inode->i_flags = ll_ext_to_inode_flags(body->flags);
1747 if (body->valid & OBD_MD_FLNLINK)
1748 set_nlink(inode, body->nlink);
1749 if (body->valid & OBD_MD_FLRDEV)
1750 inode->i_rdev = old_decode_dev(body->rdev);
1752 if (body->valid & OBD_MD_FLID) {
1753 /* FID shouldn't be changed! */
1754 if (fid_is_sane(&lli->lli_fid)) {
1755 LASSERTF(lu_fid_eq(&lli->lli_fid, &body->fid1),
1756 "Trying to change FID "DFID
1757 " to the "DFID", inode %lu/%u(%p)\n",
1758 PFID(&lli->lli_fid), PFID(&body->fid1),
1759 inode->i_ino, inode->i_generation, inode);
1761 lli->lli_fid = body->fid1;
1764 LASSERT(fid_seq(&lli->lli_fid) != 0);
1766 if (body->valid & OBD_MD_FLSIZE) {
1767 if (exp_connect_som(ll_i2mdexp(inode)) &&
1768 S_ISREG(inode->i_mode)) {
1769 struct lustre_handle lockh;
1772 /* As it is possible a blocking ast has been processed
1773 * by this time, we need to check there is an UPDATE
1774 * lock on the client and set LLIF_MDS_SIZE_LOCK holding
1776 mode = ll_take_md_lock(inode, MDS_INODELOCK_UPDATE,
1777 &lockh, LDLM_FL_CBPENDING,
1781 if (lli->lli_flags & (LLIF_DONE_WRITING |
1782 LLIF_EPOCH_PENDING |
1784 CERROR("ino %lu flags %u still has "
1785 "size authority! do not trust "
1786 "the size got from MDS\n",
1787 inode->i_ino, lli->lli_flags);
1789 /* Use old size assignment to avoid
1790 * deadlock bz14138 & bz14326 */
1791 i_size_write(inode, body->size);
1792 spin_lock(&lli->lli_lock);
1793 lli->lli_flags |= LLIF_MDS_SIZE_LOCK;
1794 spin_unlock(&lli->lli_lock);
1796 ldlm_lock_decref(&lockh, mode);
1799 /* Use old size assignment to avoid
1800 * deadlock bz14138 & bz14326 */
1801 i_size_write(inode, body->size);
1803 CDEBUG(D_VFSTRACE, "inode=%lu, updating i_size %llu\n",
1804 inode->i_ino, (unsigned long long)body->size);
1807 if (body->valid & OBD_MD_FLBLOCKS)
1808 inode->i_blocks = body->blocks;
1811 if (body->valid & OBD_MD_FLMDSCAPA) {
1812 LASSERT(md->mds_capa);
1813 ll_add_capa(inode, md->mds_capa);
1815 if (body->valid & OBD_MD_FLOSSCAPA) {
1816 LASSERT(md->oss_capa);
1817 ll_add_capa(inode, md->oss_capa);
1820 if (body->valid & OBD_MD_TSTATE) {
1821 if (body->t_state & MS_RESTORE)
1822 lli->lli_flags |= LLIF_FILE_RESTORING;
1826 void ll_read_inode2(struct inode *inode, void *opaque)
1828 struct lustre_md *md = opaque;
1829 struct ll_inode_info *lli = ll_i2info(inode);
1831 CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p)\n",
1832 PFID(&lli->lli_fid), inode);
1834 LASSERT(!lli->lli_has_smd);
1836 /* Core attributes from the MDS first. This is a new inode, and
1837 * the VFS doesn't zero times in the core inode so we have to do
1838 * it ourselves. They will be overwritten by either MDS or OST
1839 * attributes - we just need to make sure they aren't newer. */
1840 LTIME_S(inode->i_mtime) = 0;
1841 LTIME_S(inode->i_atime) = 0;
1842 LTIME_S(inode->i_ctime) = 0;
1844 ll_update_inode(inode, md);
1846 /* OIDEBUG(inode); */
1848 /* initializing backing dev info. */
1849 inode->i_mapping->backing_dev_info = &s2lsi(inode->i_sb)->lsi_bdi;
1852 if (S_ISREG(inode->i_mode)) {
1853 struct ll_sb_info *sbi = ll_i2sbi(inode);
1854 inode->i_op = &ll_file_inode_operations;
1855 inode->i_fop = sbi->ll_fop;
1856 inode->i_mapping->a_ops = (struct address_space_operations *)&ll_aops;
1857 } else if (S_ISDIR(inode->i_mode)) {
1858 inode->i_op = &ll_dir_inode_operations;
1859 inode->i_fop = &ll_dir_operations;
1860 } else if (S_ISLNK(inode->i_mode)) {
1861 inode->i_op = &ll_fast_symlink_inode_operations;
1863 inode->i_op = &ll_special_inode_operations;
1865 init_special_inode(inode, inode->i_mode,
1870 void ll_delete_inode(struct inode *inode)
1872 struct cl_inode_info *lli = cl_i2info(inode);
1874 if (S_ISREG(inode->i_mode) && lli->lli_clob != NULL)
1875 /* discard all dirty pages before truncating them, required by
1876 * osc_extent implementation at LU-1030. */
1877 cl_sync_file_range(inode, 0, OBD_OBJECT_EOF,
1878 CL_FSYNC_DISCARD, 1);
1880 truncate_inode_pages(&inode->i_data, 0);
1882 /* Workaround for LU-118 */
1883 if (inode->i_data.nrpages) {
1884 TREE_READ_LOCK_IRQ(&inode->i_data);
1885 TREE_READ_UNLOCK_IRQ(&inode->i_data);
1886 LASSERTF(inode->i_data.nrpages == 0,
1887 "inode=%lu/%u(%p) nrpages=%lu, see "
1888 "http://jira.whamcloud.com/browse/LU-118\n",
1889 inode->i_ino, inode->i_generation, inode,
1890 inode->i_data.nrpages);
1892 /* Workaround end */
1894 ll_clear_inode(inode);
1898 int ll_iocontrol(struct inode *inode, struct file *file,
1899 unsigned int cmd, unsigned long arg)
1901 struct ll_sb_info *sbi = ll_i2sbi(inode);
1902 struct ptlrpc_request *req = NULL;
1906 case FSFILT_IOC_GETFLAGS: {
1907 struct mdt_body *body;
1908 struct md_op_data *op_data;
1910 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL,
1911 0, 0, LUSTRE_OPC_ANY,
1913 if (IS_ERR(op_data))
1914 return PTR_ERR(op_data);
1916 op_data->op_valid = OBD_MD_FLFLAGS;
1917 rc = md_getattr(sbi->ll_md_exp, op_data, &req);
1918 ll_finish_md_op_data(op_data);
1920 CERROR("failure %d inode %lu\n", rc, inode->i_ino);
1924 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
1926 flags = body->flags;
1928 ptlrpc_req_finished(req);
1930 return put_user(flags, (int *)arg);
1932 case FSFILT_IOC_SETFLAGS: {
1933 struct lov_stripe_md *lsm;
1934 struct obd_info oinfo = { { { 0 } } };
1935 struct md_op_data *op_data;
1937 if (get_user(flags, (int *)arg))
1940 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
1941 LUSTRE_OPC_ANY, NULL);
1942 if (IS_ERR(op_data))
1943 return PTR_ERR(op_data);
1945 ((struct ll_iattr *)&op_data->op_attr)->ia_attr_flags = flags;
1946 op_data->op_attr.ia_valid |= ATTR_ATTR_FLAG;
1947 rc = md_setattr(sbi->ll_md_exp, op_data,
1948 NULL, 0, NULL, 0, &req, NULL);
1949 ll_finish_md_op_data(op_data);
1950 ptlrpc_req_finished(req);
1954 inode->i_flags = ll_ext_to_inode_flags(flags);
1956 lsm = ccc_inode_lsm_get(inode);
1957 if (!lsm_has_objects(lsm)) {
1958 ccc_inode_lsm_put(inode, lsm);
1962 OBDO_ALLOC(oinfo.oi_oa);
1964 ccc_inode_lsm_put(inode, lsm);
1968 oinfo.oi_oa->o_oi = lsm->lsm_oi;
1969 oinfo.oi_oa->o_flags = flags;
1970 oinfo.oi_oa->o_valid = OBD_MD_FLID | OBD_MD_FLFLAGS |
1972 oinfo.oi_capa = ll_mdscapa_get(inode);
1973 obdo_set_parent_fid(oinfo.oi_oa, &ll_i2info(inode)->lli_fid);
1974 rc = obd_setattr_rqset(sbi->ll_dt_exp, &oinfo, NULL);
1975 capa_put(oinfo.oi_capa);
1976 OBDO_FREE(oinfo.oi_oa);
1977 ccc_inode_lsm_put(inode, lsm);
1979 if (rc && rc != -EPERM && rc != -EACCES)
1980 CERROR("osc_setattr_async fails: rc = %d\n", rc);
1991 int ll_flush_ctx(struct inode *inode)
1993 struct ll_sb_info *sbi = ll_i2sbi(inode);
1995 CDEBUG(D_SEC, "flush context for user %d\n",
1996 from_kuid(&init_user_ns, current_uid()));
1998 obd_set_info_async(NULL, sbi->ll_md_exp,
1999 sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
2001 obd_set_info_async(NULL, sbi->ll_dt_exp,
2002 sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
2007 /* umount -f client means force down, don't save state */
2008 void ll_umount_begin(struct super_block *sb)
2010 struct ll_sb_info *sbi = ll_s2sbi(sb);
2011 struct obd_device *obd;
2012 struct obd_ioctl_data *ioc_data;
2014 CDEBUG(D_VFSTRACE, "VFS Op: superblock %p count %d active %d\n", sb,
2015 sb->s_count, atomic_read(&sb->s_active));
2017 obd = class_exp2obd(sbi->ll_md_exp);
2019 CERROR("Invalid MDC connection handle "LPX64"\n",
2020 sbi->ll_md_exp->exp_handle.h_cookie);
2025 obd = class_exp2obd(sbi->ll_dt_exp);
2027 CERROR("Invalid LOV connection handle "LPX64"\n",
2028 sbi->ll_dt_exp->exp_handle.h_cookie);
2033 OBD_ALLOC_PTR(ioc_data);
2035 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_md_exp,
2036 sizeof(*ioc_data), ioc_data, NULL);
2038 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_dt_exp,
2039 sizeof(*ioc_data), ioc_data, NULL);
2041 OBD_FREE_PTR(ioc_data);
2044 /* Really, we'd like to wait until there are no requests outstanding,
2045 * and then continue. For now, we just invalidate the requests,
2046 * schedule() and sleep one second if needed, and hope.
2051 int ll_remount_fs(struct super_block *sb, int *flags, char *data)
2053 struct ll_sb_info *sbi = ll_s2sbi(sb);
2054 char *profilenm = get_profile_name(sb);
2058 if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY)) {
2059 read_only = *flags & MS_RDONLY;
2060 err = obd_set_info_async(NULL, sbi->ll_md_exp,
2061 sizeof(KEY_READ_ONLY),
2062 KEY_READ_ONLY, sizeof(read_only),
2065 LCONSOLE_WARN("Failed to remount %s %s (%d)\n",
2066 profilenm, read_only ?
2067 "read-only" : "read-write", err);
2072 sb->s_flags |= MS_RDONLY;
2074 sb->s_flags &= ~MS_RDONLY;
2076 if (sbi->ll_flags & LL_SBI_VERBOSE)
2077 LCONSOLE_WARN("Remounted %s %s\n", profilenm,
2078 read_only ? "read-only" : "read-write");
2083 int ll_prep_inode(struct inode **inode, struct ptlrpc_request *req,
2084 struct super_block *sb, struct lookup_intent *it)
2086 struct ll_sb_info *sbi = NULL;
2087 struct lustre_md md;
2090 LASSERT(*inode || sb);
2091 sbi = sb ? ll_s2sbi(sb) : ll_i2sbi(*inode);
2092 rc = md_get_lustre_md(sbi->ll_md_exp, req, sbi->ll_dt_exp,
2093 sbi->ll_md_exp, &md);
2098 ll_update_inode(*inode, &md);
2100 LASSERT(sb != NULL);
2103 * At this point server returns to client's same fid as client
2104 * generated for creating. So using ->fid1 is okay here.
2106 LASSERT(fid_is_sane(&md.body->fid1));
2108 *inode = ll_iget(sb, cl_fid_build_ino(&md.body->fid1,
2109 sbi->ll_flags & LL_SBI_32BIT_API),
2111 if (*inode == NULL || IS_ERR(*inode)) {
2112 #ifdef CONFIG_FS_POSIX_ACL
2114 posix_acl_release(md.posix_acl);
2115 md.posix_acl = NULL;
2118 rc = IS_ERR(*inode) ? PTR_ERR(*inode) : -ENOMEM;
2120 CERROR("new_inode -fatal: rc %d\n", rc);
2125 /* Handling piggyback layout lock.
2126 * Layout lock can be piggybacked by getattr and open request.
2127 * The lsm can be applied to inode only if it comes with a layout lock
2128 * otherwise correct layout may be overwritten, for example:
2129 * 1. proc1: mdt returns a lsm but not granting layout
2130 * 2. layout was changed by another client
2131 * 3. proc2: refresh layout and layout lock granted
2132 * 4. proc1: to apply a stale layout */
2133 if (it != NULL && it->d.lustre.it_lock_mode != 0) {
2134 struct lustre_handle lockh;
2135 struct ldlm_lock *lock;
2137 lockh.cookie = it->d.lustre.it_lock_handle;
2138 lock = ldlm_handle2lock(&lockh);
2139 LASSERT(lock != NULL);
2140 if (ldlm_has_layout(lock)) {
2141 struct cl_object_conf conf;
2143 memset(&conf, 0, sizeof(conf));
2144 conf.coc_opc = OBJECT_CONF_SET;
2145 conf.coc_inode = *inode;
2146 conf.coc_lock = lock;
2147 conf.u.coc_md = &md;
2148 (void)ll_layout_conf(*inode, &conf);
2150 LDLM_LOCK_PUT(lock);
2155 obd_free_memmd(sbi->ll_dt_exp, &md.lsm);
2156 md_free_lustre_md(sbi->ll_md_exp, &md);
2160 int ll_obd_statfs(struct inode *inode, void *arg)
2162 struct ll_sb_info *sbi = NULL;
2163 struct obd_export *exp;
2165 struct obd_ioctl_data *data = NULL;
2170 if (!inode || !(sbi = ll_i2sbi(inode)))
2171 GOTO(out_statfs, rc = -EINVAL);
2173 rc = obd_ioctl_getdata(&buf, &len, arg);
2175 GOTO(out_statfs, rc);
2178 if (!data->ioc_inlbuf1 || !data->ioc_inlbuf2 ||
2179 !data->ioc_pbuf1 || !data->ioc_pbuf2)
2180 GOTO(out_statfs, rc = -EINVAL);
2182 if (data->ioc_inllen1 != sizeof(__u32) ||
2183 data->ioc_inllen2 != sizeof(__u32) ||
2184 data->ioc_plen1 != sizeof(struct obd_statfs) ||
2185 data->ioc_plen2 != sizeof(struct obd_uuid))
2186 GOTO(out_statfs, rc = -EINVAL);
2188 memcpy(&type, data->ioc_inlbuf1, sizeof(__u32));
2189 if (type & LL_STATFS_LMV)
2190 exp = sbi->ll_md_exp;
2191 else if (type & LL_STATFS_LOV)
2192 exp = sbi->ll_dt_exp;
2194 GOTO(out_statfs, rc = -ENODEV);
2196 flags = (type & LL_STATFS_NODELAY) ? OBD_STATFS_NODELAY : 0;
2197 rc = obd_iocontrol(IOC_OBD_STATFS, exp, len, buf, &flags);
2199 GOTO(out_statfs, rc);
2202 obd_ioctl_freedata(buf, len);
2206 int ll_process_config(struct lustre_cfg *lcfg)
2210 struct lprocfs_static_vars lvars;
2214 lprocfs_llite_init_vars(&lvars);
2216 /* The instance name contains the sb: lustre-client-aacfe000 */
2217 ptr = strrchr(lustre_cfg_string(lcfg, 0), '-');
2218 if (!ptr || !*(++ptr))
2220 if (sscanf(ptr, "%lx", &x) != 1)
2223 /* This better be a real Lustre superblock! */
2224 LASSERT(s2lsi((struct super_block *)sb)->lsi_lmd->lmd_magic == LMD_MAGIC);
2226 /* Note we have not called client_common_fill_super yet, so
2227 proc fns must be able to handle that! */
2228 rc = class_process_proc_param(PARAM_LLITE, lvars.obd_vars,
2235 /* this function prepares md_op_data hint for passing ot down to MD stack. */
2236 struct md_op_data * ll_prep_md_op_data(struct md_op_data *op_data,
2237 struct inode *i1, struct inode *i2,
2238 const char *name, int namelen,
2239 int mode, __u32 opc, void *data)
2241 LASSERT(i1 != NULL);
2243 if (namelen > ll_i2sbi(i1)->ll_namelen)
2244 return ERR_PTR(-ENAMETOOLONG);
2246 if (op_data == NULL)
2247 OBD_ALLOC_PTR(op_data);
2249 if (op_data == NULL)
2250 return ERR_PTR(-ENOMEM);
2252 ll_i2gids(op_data->op_suppgids, i1, i2);
2253 op_data->op_fid1 = *ll_inode2fid(i1);
2254 op_data->op_capa1 = ll_mdscapa_get(i1);
2257 op_data->op_fid2 = *ll_inode2fid(i2);
2258 op_data->op_capa2 = ll_mdscapa_get(i2);
2260 fid_zero(&op_data->op_fid2);
2261 op_data->op_capa2 = NULL;
2264 op_data->op_name = name;
2265 op_data->op_namelen = namelen;
2266 op_data->op_mode = mode;
2267 op_data->op_mod_time = cfs_time_current_sec();
2268 op_data->op_fsuid = from_kuid(&init_user_ns, current_fsuid());
2269 op_data->op_fsgid = from_kgid(&init_user_ns, current_fsgid());
2270 op_data->op_cap = cfs_curproc_cap_pack();
2271 op_data->op_bias = 0;
2272 op_data->op_cli_flags = 0;
2273 if ((opc == LUSTRE_OPC_CREATE) && (name != NULL) &&
2274 filename_is_volatile(name, namelen, NULL))
2275 op_data->op_bias |= MDS_CREATE_VOLATILE;
2276 op_data->op_opc = opc;
2277 op_data->op_mds = 0;
2278 op_data->op_data = data;
2280 /* If the file is being opened after mknod() (normally due to NFS)
2281 * try to use the default stripe data from parent directory for
2282 * allocating OST objects. Try to pass the parent FID to MDS. */
2283 if (opc == LUSTRE_OPC_CREATE && i1 == i2 && S_ISREG(i2->i_mode) &&
2284 !ll_i2info(i2)->lli_has_smd) {
2285 struct ll_inode_info *lli = ll_i2info(i2);
2287 spin_lock(&lli->lli_lock);
2288 if (likely(!lli->lli_has_smd && !fid_is_zero(&lli->lli_pfid)))
2289 op_data->op_fid1 = lli->lli_pfid;
2290 spin_unlock(&lli->lli_lock);
2291 /** We ignore parent's capability temporary. */
2294 /* When called by ll_setattr_raw, file is i1. */
2295 if (LLIF_DATA_MODIFIED & ll_i2info(i1)->lli_flags)
2296 op_data->op_bias |= MDS_DATA_MODIFIED;
2301 void ll_finish_md_op_data(struct md_op_data *op_data)
2303 capa_put(op_data->op_capa1);
2304 capa_put(op_data->op_capa2);
2305 OBD_FREE_PTR(op_data);
2308 int ll_show_options(struct seq_file *seq, struct dentry *dentry)
2310 struct ll_sb_info *sbi;
2312 LASSERT((seq != NULL) && (dentry != NULL));
2313 sbi = ll_s2sbi(dentry->d_sb);
2315 if (sbi->ll_flags & LL_SBI_NOLCK)
2316 seq_puts(seq, ",nolock");
2318 if (sbi->ll_flags & LL_SBI_FLOCK)
2319 seq_puts(seq, ",flock");
2321 if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
2322 seq_puts(seq, ",localflock");
2324 if (sbi->ll_flags & LL_SBI_USER_XATTR)
2325 seq_puts(seq, ",user_xattr");
2327 if (sbi->ll_flags & LL_SBI_LAZYSTATFS)
2328 seq_puts(seq, ",lazystatfs");
2330 if (sbi->ll_flags & LL_SBI_USER_FID2PATH)
2331 seq_puts(seq, ",user_fid2path");
2337 * Get obd name by cmd, and copy out to user space
2339 int ll_get_obd_name(struct inode *inode, unsigned int cmd, unsigned long arg)
2341 struct ll_sb_info *sbi = ll_i2sbi(inode);
2342 struct obd_device *obd;
2344 if (cmd == OBD_IOC_GETDTNAME)
2345 obd = class_exp2obd(sbi->ll_dt_exp);
2346 else if (cmd == OBD_IOC_GETMDNAME)
2347 obd = class_exp2obd(sbi->ll_md_exp);
2354 if (copy_to_user((void *)arg, obd->obd_name,
2355 strlen(obd->obd_name) + 1))
2362 * Get lustre file system name by \a sbi. If \a buf is provided(non-NULL), the
2363 * fsname will be returned in this buffer; otherwise, a static buffer will be
2364 * used to store the fsname and returned to caller.
2366 char *ll_get_fsname(struct super_block *sb, char *buf, int buflen)
2368 static char fsname_static[MTI_NAME_MAXLEN];
2369 struct lustre_sb_info *lsi = s2lsi(sb);
2374 /* this means the caller wants to use static buffer
2375 * and it doesn't care about race. Usually this is
2376 * in error reporting path */
2377 buf = fsname_static;
2378 buflen = sizeof(fsname_static);
2381 len = strlen(lsi->lsi_lmd->lmd_profile);
2382 ptr = strrchr(lsi->lsi_lmd->lmd_profile, '-');
2383 if (ptr && (strcmp(ptr, "-client") == 0))
2386 if (unlikely(len >= buflen))
2388 strncpy(buf, lsi->lsi_lmd->lmd_profile, len);
2394 static char* ll_d_path(struct dentry *dentry, char *buf, int bufsize)
2401 p.mnt = current->fs->root.mnt;
2403 path = d_path(&p, buf, bufsize);
2409 void ll_dirty_page_discard_warn(struct page *page, int ioret)
2411 char *buf, *path = NULL;
2412 struct dentry *dentry = NULL;
2413 struct ccc_object *obj = cl_inode2ccc(page->mapping->host);
2415 /* this can be called inside spin lock so use GFP_ATOMIC. */
2416 buf = (char *)__get_free_page(GFP_ATOMIC);
2418 dentry = d_find_alias(page->mapping->host);
2420 path = ll_d_path(dentry, buf, PAGE_SIZE);
2423 CWARN("%s: dirty page discard: %s/fid: "DFID"/%s may get corrupted "
2424 "(rc %d)\n", ll_get_fsname(page->mapping->host->i_sb, NULL, 0),
2425 s2lsi(page->mapping->host->i_sb)->lsi_lmd->lmd_dev,
2426 PFID(&obj->cob_header.coh_lu.loh_fid),
2427 (path && !IS_ERR(path)) ? path : "", ioret);
2433 free_page((unsigned long)buf);