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;
60 LIST_HEAD(ll_super_blocks);
61 DEFINE_SPINLOCK(ll_sb_lock);
63 #ifndef MS_HAS_NEW_AOPS
64 extern struct address_space_operations ll_aops;
66 extern struct address_space_operations_ext ll_aops;
70 #define log2(n) ffz(~(n))
73 static struct ll_sb_info *ll_init_sbi(void)
75 struct ll_sb_info *sbi = NULL;
77 unsigned long lru_page_max;
82 OBD_ALLOC(sbi, sizeof(*sbi));
86 spin_lock_init(&sbi->ll_lock);
87 mutex_init(&sbi->ll_lco.lco_lock);
88 spin_lock_init(&sbi->ll_pp_extent_lock);
89 spin_lock_init(&sbi->ll_process_lock);
90 sbi->ll_rw_stats_on = 0;
93 pages = si.totalram - si.totalhigh;
94 if (pages >> (20 - PAGE_CACHE_SHIFT) < 512) {
95 lru_page_max = pages / 2;
97 lru_page_max = (pages / 4) * 3;
100 /* initialize lru data */
101 atomic_set(&sbi->ll_cache.ccc_users, 0);
102 sbi->ll_cache.ccc_lru_max = lru_page_max;
103 atomic_set(&sbi->ll_cache.ccc_lru_left, lru_page_max);
104 spin_lock_init(&sbi->ll_cache.ccc_lru_lock);
105 INIT_LIST_HEAD(&sbi->ll_cache.ccc_lru);
107 sbi->ll_ra_info.ra_max_pages_per_file = min(pages / 32,
108 SBI_DEFAULT_READAHEAD_MAX);
109 sbi->ll_ra_info.ra_max_pages = sbi->ll_ra_info.ra_max_pages_per_file;
110 sbi->ll_ra_info.ra_max_read_ahead_whole_pages =
111 SBI_DEFAULT_READAHEAD_WHOLE_MAX;
112 INIT_LIST_HEAD(&sbi->ll_conn_chain);
113 INIT_LIST_HEAD(&sbi->ll_orphan_dentry_list);
115 ll_generate_random_uuid(uuid);
116 class_uuid_unparse(uuid, &sbi->ll_sb_uuid);
117 CDEBUG(D_CONFIG, "generated uuid: %s\n", sbi->ll_sb_uuid.uuid);
119 spin_lock(&ll_sb_lock);
120 list_add_tail(&sbi->ll_list, &ll_super_blocks);
121 spin_unlock(&ll_sb_lock);
123 sbi->ll_flags |= LL_SBI_VERBOSE;
124 sbi->ll_flags |= LL_SBI_CHECKSUM;
126 sbi->ll_flags |= LL_SBI_LRU_RESIZE;
128 for (i = 0; i <= LL_PROCESS_HIST_MAX; i++) {
129 spin_lock_init(&sbi->ll_rw_extents_info.pp_extents[i].
131 spin_lock_init(&sbi->ll_rw_extents_info.pp_extents[i].
135 /* metadata statahead is enabled by default */
136 sbi->ll_sa_max = LL_SA_RPC_DEF;
137 atomic_set(&sbi->ll_sa_total, 0);
138 atomic_set(&sbi->ll_sa_wrong, 0);
139 atomic_set(&sbi->ll_agl_total, 0);
140 sbi->ll_flags |= LL_SBI_AGL_ENABLED;
145 void ll_free_sbi(struct super_block *sb)
147 struct ll_sb_info *sbi = ll_s2sbi(sb);
150 spin_lock(&ll_sb_lock);
151 list_del(&sbi->ll_list);
152 spin_unlock(&ll_sb_lock);
153 OBD_FREE(sbi, sizeof(*sbi));
157 static struct dentry_operations ll_d_root_ops = {
158 .d_compare = ll_dcompare,
159 .d_revalidate = ll_revalidate_nd,
162 static int client_common_fill_super(struct super_block *sb, char *md, char *dt,
163 struct vfsmount *mnt)
165 struct inode *root = 0;
166 struct ll_sb_info *sbi = ll_s2sbi(sb);
167 struct obd_device *obd;
168 struct obd_capa *oc = NULL;
169 struct obd_statfs *osfs = NULL;
170 struct ptlrpc_request *request = NULL;
171 struct obd_connect_data *data = NULL;
172 struct obd_uuid *uuid;
173 struct md_op_data *op_data;
174 struct lustre_md lmd;
176 int size, err, checksum;
178 obd = class_name2obd(md);
180 CERROR("MD %s: not setup or attached\n", md);
194 if (proc_lustre_fs_root) {
195 err = lprocfs_register_mountpoint(proc_lustre_fs_root, sb,
198 CERROR("could not register mount in /proc/fs/lustre\n");
201 /* indicate the features supported by this client */
202 data->ocd_connect_flags = OBD_CONNECT_IBITS | OBD_CONNECT_NODEVOH |
203 OBD_CONNECT_ATTRFID |
204 OBD_CONNECT_VERSION | OBD_CONNECT_BRW_SIZE |
205 OBD_CONNECT_MDS_CAPA | OBD_CONNECT_OSS_CAPA |
206 OBD_CONNECT_CANCELSET | OBD_CONNECT_FID |
207 OBD_CONNECT_AT | OBD_CONNECT_LOV_V3 |
208 OBD_CONNECT_RMT_CLIENT | OBD_CONNECT_VBR |
209 OBD_CONNECT_FULL20 | OBD_CONNECT_64BITHASH|
210 OBD_CONNECT_EINPROGRESS |
211 OBD_CONNECT_JOBSTATS | OBD_CONNECT_LVB_TYPE |
212 OBD_CONNECT_LAYOUTLOCK | OBD_CONNECT_PINGLESS;
214 if (sbi->ll_flags & LL_SBI_SOM_PREVIEW)
215 data->ocd_connect_flags |= OBD_CONNECT_SOM;
217 if (sbi->ll_flags & LL_SBI_LRU_RESIZE)
218 data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
219 #ifdef CONFIG_FS_POSIX_ACL
220 data->ocd_connect_flags |= OBD_CONNECT_ACL | OBD_CONNECT_UMASK;
223 if (OBD_FAIL_CHECK(OBD_FAIL_MDC_LIGHTWEIGHT))
224 /* flag mdc connection as lightweight, only used for test
225 * purpose, use with care */
226 data->ocd_connect_flags |= OBD_CONNECT_LIGHTWEIGHT;
228 data->ocd_ibits_known = MDS_INODELOCK_FULL;
229 data->ocd_version = LUSTRE_VERSION_CODE;
231 if (sb->s_flags & MS_RDONLY)
232 data->ocd_connect_flags |= OBD_CONNECT_RDONLY;
233 if (sbi->ll_flags & LL_SBI_USER_XATTR)
234 data->ocd_connect_flags |= OBD_CONNECT_XATTR;
236 #ifdef HAVE_MS_FLOCK_LOCK
237 /* force vfs to use lustre handler for flock() calls - bug 10743 */
238 sb->s_flags |= MS_FLOCK_LOCK;
240 #ifdef MS_HAS_NEW_AOPS
241 sb->s_flags |= MS_HAS_NEW_AOPS;
244 if (sbi->ll_flags & LL_SBI_FLOCK)
245 sbi->ll_fop = &ll_file_operations_flock;
246 else if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
247 sbi->ll_fop = &ll_file_operations;
249 sbi->ll_fop = &ll_file_operations_noflock;
252 data->ocd_connect_flags |= OBD_CONNECT_REAL;
253 if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
254 data->ocd_connect_flags |= OBD_CONNECT_RMT_CLIENT_FORCE;
256 data->ocd_brw_size = MD_MAX_BRW_SIZE;
258 err = obd_connect(NULL, &sbi->ll_md_exp, obd, &sbi->ll_sb_uuid, data, NULL);
260 LCONSOLE_ERROR_MSG(0x14f, "An MDT (md %s) is performing "
261 "recovery, of which this client is not a "
262 "part. Please wait for recovery to complete,"
263 " abort, or time out.\n", md);
266 CERROR("cannot connect to %s: rc = %d\n", md, err);
270 sbi->ll_md_exp->exp_connect_data = *data;
272 err = obd_fid_init(sbi->ll_md_exp->exp_obd, sbi->ll_md_exp,
273 LUSTRE_SEQ_METADATA);
275 CERROR("%s: Can't init metadata layer FID infrastructure, "
276 "rc = %d\n", sbi->ll_md_exp->exp_obd->obd_name, err);
280 /* For mount, we only need fs info from MDT0, and also in DNE, it
281 * can make sure the client can be mounted as long as MDT0 is
283 err = obd_statfs(NULL, sbi->ll_md_exp, osfs,
284 cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS),
285 OBD_STATFS_FOR_MDT0);
287 GOTO(out_md_fid, err);
289 /* This needs to be after statfs to ensure connect has finished.
290 * Note that "data" does NOT contain the valid connect reply.
291 * If connecting to a 1.8 server there will be no LMV device, so
292 * we can access the MDC export directly and exp_connect_flags will
293 * be non-zero, but if accessing an upgraded 2.1 server it will
294 * have the correct flags filled in.
295 * XXX: fill in the LMV exp_connect_flags from MDC(s). */
296 valid = exp_connect_flags(sbi->ll_md_exp) & CLIENT_CONNECT_MDT_REQD;
297 if (exp_connect_flags(sbi->ll_md_exp) != 0 &&
298 valid != CLIENT_CONNECT_MDT_REQD) {
301 OBD_ALLOC_WAIT(buf, PAGE_CACHE_SIZE);
302 obd_connect_flags2str(buf, PAGE_CACHE_SIZE,
303 valid ^ CLIENT_CONNECT_MDT_REQD, ",");
304 LCONSOLE_ERROR_MSG(0x170, "Server %s does not support "
305 "feature(s) needed for correct operation "
306 "of this client (%s). Please upgrade "
307 "server or downgrade client.\n",
308 sbi->ll_md_exp->exp_obd->obd_name, buf);
309 OBD_FREE(buf, PAGE_CACHE_SIZE);
310 GOTO(out_md_fid, err = -EPROTO);
313 size = sizeof(*data);
314 err = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_CONN_DATA),
315 KEY_CONN_DATA, &size, data, NULL);
317 CERROR("%s: Get connect data failed: rc = %d\n",
318 sbi->ll_md_exp->exp_obd->obd_name, err);
319 GOTO(out_md_fid, err);
322 LASSERT(osfs->os_bsize);
323 sb->s_blocksize = osfs->os_bsize;
324 sb->s_blocksize_bits = log2(osfs->os_bsize);
325 sb->s_magic = LL_SUPER_MAGIC;
326 sb->s_maxbytes = MAX_LFS_FILESIZE;
327 sbi->ll_namelen = osfs->os_namelen;
328 sbi->ll_max_rw_chunk = LL_DEFAULT_MAX_RW_CHUNK;
330 if ((sbi->ll_flags & LL_SBI_USER_XATTR) &&
331 !(data->ocd_connect_flags & OBD_CONNECT_XATTR)) {
332 LCONSOLE_INFO("Disabling user_xattr feature because "
333 "it is not supported on the server\n");
334 sbi->ll_flags &= ~LL_SBI_USER_XATTR;
337 if (data->ocd_connect_flags & OBD_CONNECT_ACL) {
339 sb->s_flags |= MS_POSIXACL;
341 sbi->ll_flags |= LL_SBI_ACL;
343 LCONSOLE_INFO("client wants to enable acl, but mdt not!\n");
345 sb->s_flags &= ~MS_POSIXACL;
347 sbi->ll_flags &= ~LL_SBI_ACL;
350 if (data->ocd_connect_flags & OBD_CONNECT_RMT_CLIENT) {
351 if (!(sbi->ll_flags & LL_SBI_RMT_CLIENT)) {
352 sbi->ll_flags |= LL_SBI_RMT_CLIENT;
353 LCONSOLE_INFO("client is set as remote by default.\n");
356 if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
357 sbi->ll_flags &= ~LL_SBI_RMT_CLIENT;
358 LCONSOLE_INFO("client claims to be remote, but server "
359 "rejected, forced to be local.\n");
363 if (data->ocd_connect_flags & OBD_CONNECT_MDS_CAPA) {
364 LCONSOLE_INFO("client enabled MDS capability!\n");
365 sbi->ll_flags |= LL_SBI_MDS_CAPA;
368 if (data->ocd_connect_flags & OBD_CONNECT_OSS_CAPA) {
369 LCONSOLE_INFO("client enabled OSS capability!\n");
370 sbi->ll_flags |= LL_SBI_OSS_CAPA;
373 if (data->ocd_connect_flags & OBD_CONNECT_64BITHASH)
374 sbi->ll_flags |= LL_SBI_64BIT_HASH;
376 if (data->ocd_connect_flags & OBD_CONNECT_BRW_SIZE)
377 sbi->ll_md_brw_size = data->ocd_brw_size;
379 sbi->ll_md_brw_size = PAGE_CACHE_SIZE;
381 if (data->ocd_connect_flags & OBD_CONNECT_LAYOUTLOCK) {
382 LCONSOLE_INFO("Layout lock feature supported.\n");
383 sbi->ll_flags |= LL_SBI_LAYOUT_LOCK;
386 obd = class_name2obd(dt);
388 CERROR("DT %s: not setup or attached\n", dt);
389 GOTO(out_md_fid, err = -ENODEV);
392 data->ocd_connect_flags = OBD_CONNECT_GRANT | OBD_CONNECT_VERSION |
393 OBD_CONNECT_REQPORTAL | OBD_CONNECT_BRW_SIZE |
394 OBD_CONNECT_CANCELSET | OBD_CONNECT_FID |
395 OBD_CONNECT_SRVLOCK | OBD_CONNECT_TRUNCLOCK|
396 OBD_CONNECT_AT | OBD_CONNECT_RMT_CLIENT |
397 OBD_CONNECT_OSS_CAPA | OBD_CONNECT_VBR|
398 OBD_CONNECT_FULL20 | OBD_CONNECT_64BITHASH |
399 OBD_CONNECT_MAXBYTES |
400 OBD_CONNECT_EINPROGRESS |
401 OBD_CONNECT_JOBSTATS | OBD_CONNECT_LVB_TYPE |
402 OBD_CONNECT_LAYOUTLOCK | OBD_CONNECT_PINGLESS;
404 if (sbi->ll_flags & LL_SBI_SOM_PREVIEW)
405 data->ocd_connect_flags |= OBD_CONNECT_SOM;
407 if (!OBD_FAIL_CHECK(OBD_FAIL_OSC_CONNECT_CKSUM)) {
408 /* OBD_CONNECT_CKSUM should always be set, even if checksums are
409 * disabled by default, because it can still be enabled on the
410 * fly via /proc. As a consequence, we still need to come to an
411 * agreement on the supported algorithms at connect time */
412 data->ocd_connect_flags |= OBD_CONNECT_CKSUM;
414 if (OBD_FAIL_CHECK(OBD_FAIL_OSC_CKSUM_ADLER_ONLY))
415 data->ocd_cksum_types = OBD_CKSUM_ADLER;
417 data->ocd_cksum_types = cksum_types_supported_client();
420 data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
421 if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
422 data->ocd_connect_flags |= OBD_CONNECT_RMT_CLIENT_FORCE;
424 CDEBUG(D_RPCTRACE, "ocd_connect_flags: "LPX64" ocd_version: %d "
425 "ocd_grant: %d\n", data->ocd_connect_flags,
426 data->ocd_version, data->ocd_grant);
428 obd->obd_upcall.onu_owner = &sbi->ll_lco;
429 obd->obd_upcall.onu_upcall = cl_ocd_update;
431 data->ocd_brw_size = DT_MAX_BRW_SIZE;
433 err = obd_connect(NULL, &sbi->ll_dt_exp, obd, &sbi->ll_sb_uuid, data,
436 LCONSOLE_ERROR_MSG(0x150, "An OST (dt %s) is performing "
437 "recovery, of which this client is not a "
438 "part. Please wait for recovery to "
439 "complete, abort, or time out.\n", dt);
442 CERROR("%s: Cannot connect to %s: rc = %d\n",
443 sbi->ll_dt_exp->exp_obd->obd_name, dt, err);
447 sbi->ll_dt_exp->exp_connect_data = *data;
449 err = obd_fid_init(sbi->ll_dt_exp->exp_obd, sbi->ll_dt_exp,
450 LUSTRE_SEQ_METADATA);
452 CERROR("%s: Can't init data layer FID infrastructure, "
453 "rc = %d\n", sbi->ll_dt_exp->exp_obd->obd_name, err);
457 mutex_lock(&sbi->ll_lco.lco_lock);
458 sbi->ll_lco.lco_flags = data->ocd_connect_flags;
459 sbi->ll_lco.lco_md_exp = sbi->ll_md_exp;
460 sbi->ll_lco.lco_dt_exp = sbi->ll_dt_exp;
461 mutex_unlock(&sbi->ll_lco.lco_lock);
463 fid_zero(&sbi->ll_root_fid);
464 err = md_getstatus(sbi->ll_md_exp, &sbi->ll_root_fid, &oc);
466 CERROR("cannot mds_connect: rc = %d\n", err);
467 GOTO(out_lock_cn_cb, err);
469 if (!fid_is_sane(&sbi->ll_root_fid)) {
470 CERROR("%s: Invalid root fid "DFID" during mount\n",
471 sbi->ll_md_exp->exp_obd->obd_name,
472 PFID(&sbi->ll_root_fid));
473 GOTO(out_lock_cn_cb, err = -EINVAL);
475 CDEBUG(D_SUPER, "rootfid "DFID"\n", PFID(&sbi->ll_root_fid));
477 sb->s_op = &lustre_super_operations;
478 #if THREAD_SIZE >= 8192 /*b=17630*/
479 sb->s_export_op = &lustre_export_operations;
483 * XXX: move this to after cbd setup? */
484 valid = OBD_MD_FLGETATTR | OBD_MD_FLBLOCKS | OBD_MD_FLMDSCAPA;
485 if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
486 valid |= OBD_MD_FLRMTPERM;
487 else if (sbi->ll_flags & LL_SBI_ACL)
488 valid |= OBD_MD_FLACL;
490 OBD_ALLOC_PTR(op_data);
492 GOTO(out_lock_cn_cb, err = -ENOMEM);
494 op_data->op_fid1 = sbi->ll_root_fid;
495 op_data->op_mode = 0;
496 op_data->op_capa1 = oc;
497 op_data->op_valid = valid;
499 err = md_getattr(sbi->ll_md_exp, op_data, &request);
502 OBD_FREE_PTR(op_data);
504 CERROR("%s: md_getattr failed for root: rc = %d\n",
505 sbi->ll_md_exp->exp_obd->obd_name, err);
506 GOTO(out_lock_cn_cb, err);
509 err = md_get_lustre_md(sbi->ll_md_exp, request, sbi->ll_dt_exp,
510 sbi->ll_md_exp, &lmd);
512 CERROR("failed to understand root inode md: rc = %d\n", err);
513 ptlrpc_req_finished(request);
514 GOTO(out_lock_cn_cb, err);
517 LASSERT(fid_is_sane(&sbi->ll_root_fid));
518 root = ll_iget(sb, cl_fid_build_ino(&sbi->ll_root_fid,
519 sbi->ll_flags & LL_SBI_32BIT_API),
521 md_free_lustre_md(sbi->ll_md_exp, &lmd);
522 ptlrpc_req_finished(request);
524 if (root == NULL || IS_ERR(root)) {
526 obd_free_memmd(sbi->ll_dt_exp, &lmd.lsm);
527 #ifdef CONFIG_FS_POSIX_ACL
529 posix_acl_release(lmd.posix_acl);
530 lmd.posix_acl = NULL;
533 err = IS_ERR(root) ? PTR_ERR(root) : -EBADF;
535 CERROR("lustre_lite: bad iget4 for root\n");
539 err = ll_close_thread_start(&sbi->ll_lcq);
541 CERROR("cannot start close thread: rc %d\n", err);
545 #ifdef CONFIG_FS_POSIX_ACL
546 if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
547 rct_init(&sbi->ll_rct);
548 et_init(&sbi->ll_et);
552 checksum = sbi->ll_flags & LL_SBI_CHECKSUM;
553 err = obd_set_info_async(NULL, sbi->ll_dt_exp, sizeof(KEY_CHECKSUM),
554 KEY_CHECKSUM, sizeof(checksum), &checksum,
558 err = obd_set_info_async(NULL, sbi->ll_dt_exp, sizeof(KEY_CACHE_SET),
559 KEY_CACHE_SET, sizeof(sbi->ll_cache),
560 &sbi->ll_cache, NULL);
562 sb->s_root = d_make_root(root);
563 if (sb->s_root == NULL) {
564 CERROR("%s: can't make root dentry\n",
565 ll_get_fsname(sb, NULL, 0));
566 GOTO(out_root, err = -ENOMEM);
569 /* kernel >= 2.6.38 store dentry operations in sb->s_d_op. */
570 d_set_d_op(sb->s_root, &ll_d_root_ops);
571 sb->s_d_op = &ll_d_ops;
573 sbi->ll_sdev_orig = sb->s_dev;
575 /* We set sb->s_dev equal on all lustre clients in order to support
576 * NFS export clustering. NFSD requires that the FSID be the same
578 /* s_dev is also used in lt_compare() to compare two fs, but that is
579 * only a node-local comparison. */
580 uuid = obd_get_uuid(sbi->ll_md_exp);
582 sb->s_dev = get_uuid2int(uuid->uuid, strlen(uuid->uuid));
583 get_uuid2fsid(uuid->uuid, strlen(uuid->uuid), &sbi->ll_fsid);
596 obd_fid_fini(sbi->ll_dt_exp->exp_obd);
598 obd_disconnect(sbi->ll_dt_exp);
599 sbi->ll_dt_exp = NULL;
600 /* Make sure all OScs are gone, since cl_cache is accessing sbi. */
601 obd_zombie_barrier();
603 obd_fid_fini(sbi->ll_md_exp->exp_obd);
605 obd_disconnect(sbi->ll_md_exp);
606 sbi->ll_md_exp = NULL;
612 lprocfs_unregister_mountpoint(sbi);
616 int ll_get_max_mdsize(struct ll_sb_info *sbi, int *lmmsize)
620 *lmmsize = obd_size_diskmd(sbi->ll_dt_exp, NULL);
622 rc = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_MAX_EASIZE),
623 KEY_MAX_EASIZE, &size, lmmsize, NULL);
625 CERROR("Get max mdsize error rc %d \n", rc);
630 void ll_dump_inode(struct inode *inode)
632 struct ll_d_hlist_node *tmp;
633 int dentry_count = 0;
635 LASSERT(inode != NULL);
637 ll_d_hlist_for_each(tmp, &inode->i_dentry)
640 CERROR("inode %p dump: dev=%s ino=%lu mode=%o count=%u, %d dentries\n",
641 inode, ll_i2mdexp(inode)->exp_obd->obd_name, inode->i_ino,
642 inode->i_mode, atomic_read(&inode->i_count), dentry_count);
645 void lustre_dump_dentry(struct dentry *dentry, int recur)
647 struct list_head *tmp;
650 LASSERT(dentry != NULL);
652 list_for_each(tmp, &dentry->d_subdirs)
655 CERROR("dentry %p dump: name=%.*s parent=%.*s (%p), inode=%p, count=%u,"
656 " flags=0x%x, fsdata=%p, %d subdirs\n", dentry,
657 dentry->d_name.len, dentry->d_name.name,
658 dentry->d_parent->d_name.len, dentry->d_parent->d_name.name,
659 dentry->d_parent, dentry->d_inode, d_count(dentry),
660 dentry->d_flags, dentry->d_fsdata, subdirs);
661 if (dentry->d_inode != NULL)
662 ll_dump_inode(dentry->d_inode);
667 list_for_each(tmp, &dentry->d_subdirs) {
668 struct dentry *d = list_entry(tmp, struct dentry, d_u.d_child);
669 lustre_dump_dentry(d, recur - 1);
673 void client_common_put_super(struct super_block *sb)
675 struct ll_sb_info *sbi = ll_s2sbi(sb);
677 #ifdef CONFIG_FS_POSIX_ACL
678 if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
679 et_fini(&sbi->ll_et);
680 rct_fini(&sbi->ll_rct);
684 ll_close_thread_shutdown(sbi->ll_lcq);
688 list_del(&sbi->ll_conn_chain);
690 obd_fid_fini(sbi->ll_dt_exp->exp_obd);
691 obd_disconnect(sbi->ll_dt_exp);
692 sbi->ll_dt_exp = NULL;
693 /* wait till all OSCs are gone, since cl_cache is accessing sbi.
695 obd_zombie_barrier();
697 lprocfs_unregister_mountpoint(sbi);
699 obd_fid_fini(sbi->ll_md_exp->exp_obd);
700 obd_disconnect(sbi->ll_md_exp);
701 sbi->ll_md_exp = NULL;
704 void ll_kill_super(struct super_block *sb)
706 struct ll_sb_info *sbi;
709 if (!(sb->s_flags & MS_ACTIVE))
713 /* we need restore s_dev from changed for clustred NFS before put_super
714 * because new kernels have cached s_dev and change sb->s_dev in
715 * put_super not affected real removing devices */
717 sb->s_dev = sbi->ll_sdev_orig;
718 sbi->ll_umounting = 1;
722 char *ll_read_opt(const char *opt, char *data)
727 CDEBUG(D_SUPER, "option: %s, data %s\n", opt, data);
728 if (strncmp(opt, data, strlen(opt)))
730 if ((value = strchr(data, '=')) == NULL)
734 OBD_ALLOC(retval, strlen(value) + 1);
736 CERROR("out of memory!\n");
740 memcpy(retval, value, strlen(value)+1);
741 CDEBUG(D_SUPER, "Assigned option: %s, value %s\n", opt, retval);
745 static inline int ll_set_opt(const char *opt, char *data, int fl)
747 if (strncmp(opt, data, strlen(opt)) != 0)
753 /* non-client-specific mount options are parsed in lmd_parse */
754 static int ll_options(char *options, int *flags)
757 char *s1 = options, *s2;
762 CDEBUG(D_CONFIG, "Parsing opts %s\n", options);
765 CDEBUG(D_SUPER, "next opt=%s\n", s1);
766 tmp = ll_set_opt("nolock", s1, LL_SBI_NOLCK);
771 tmp = ll_set_opt("flock", s1, LL_SBI_FLOCK);
776 tmp = ll_set_opt("localflock", s1, LL_SBI_LOCALFLOCK);
781 tmp = ll_set_opt("noflock", s1, LL_SBI_FLOCK|LL_SBI_LOCALFLOCK);
786 tmp = ll_set_opt("user_xattr", s1, LL_SBI_USER_XATTR);
791 tmp = ll_set_opt("nouser_xattr", s1, LL_SBI_USER_XATTR);
796 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 5, 50, 0)
797 tmp = ll_set_opt("acl", s1, LL_SBI_ACL);
799 /* Ignore deprecated mount option. The client will
800 * always try to mount with ACL support, whether this
801 * is used depends on whether server supports it. */
802 LCONSOLE_ERROR_MSG(0x152, "Ignoring deprecated "
803 "mount option 'acl'.\n");
806 tmp = ll_set_opt("noacl", s1, LL_SBI_ACL);
808 LCONSOLE_ERROR_MSG(0x152, "Ignoring deprecated "
809 "mount option 'noacl'.\n");
813 #warning "{no}acl options have been deprecated since 1.8, please remove them"
815 tmp = ll_set_opt("remote_client", s1, LL_SBI_RMT_CLIENT);
820 tmp = ll_set_opt("user_fid2path", s1, LL_SBI_USER_FID2PATH);
825 tmp = ll_set_opt("nouser_fid2path", s1, LL_SBI_USER_FID2PATH);
831 tmp = ll_set_opt("checksum", s1, LL_SBI_CHECKSUM);
836 tmp = ll_set_opt("nochecksum", s1, LL_SBI_CHECKSUM);
841 tmp = ll_set_opt("lruresize", s1, LL_SBI_LRU_RESIZE);
846 tmp = ll_set_opt("nolruresize", s1, LL_SBI_LRU_RESIZE);
851 tmp = ll_set_opt("lazystatfs", s1, LL_SBI_LAZYSTATFS);
856 tmp = ll_set_opt("nolazystatfs", s1, LL_SBI_LAZYSTATFS);
861 tmp = ll_set_opt("som_preview", s1, LL_SBI_SOM_PREVIEW);
866 tmp = ll_set_opt("32bitapi", s1, LL_SBI_32BIT_API);
871 tmp = ll_set_opt("verbose", s1, LL_SBI_VERBOSE);
876 tmp = ll_set_opt("noverbose", s1, LL_SBI_VERBOSE);
881 LCONSOLE_ERROR_MSG(0x152, "Unknown option '%s', won't mount.\n",
887 s2 = strchr(s1, ',');
895 void ll_lli_init(struct ll_inode_info *lli)
897 lli->lli_inode_magic = LLI_INODE_MAGIC;
899 lli->lli_ioepoch = 0;
900 lli->lli_maxbytes = MAX_LFS_FILESIZE;
901 spin_lock_init(&lli->lli_lock);
902 lli->lli_posix_acl = NULL;
903 lli->lli_remote_perms = NULL;
904 mutex_init(&lli->lli_rmtperm_mutex);
905 /* Do not set lli_fid, it has been initialized already. */
906 fid_zero(&lli->lli_pfid);
907 INIT_LIST_HEAD(&lli->lli_close_list);
908 INIT_LIST_HEAD(&lli->lli_oss_capas);
909 atomic_set(&lli->lli_open_count, 0);
910 lli->lli_mds_capa = NULL;
911 lli->lli_rmtperm_time = 0;
912 lli->lli_pending_och = NULL;
913 lli->lli_mds_read_och = NULL;
914 lli->lli_mds_write_och = NULL;
915 lli->lli_mds_exec_och = NULL;
916 lli->lli_open_fd_read_count = 0;
917 lli->lli_open_fd_write_count = 0;
918 lli->lli_open_fd_exec_count = 0;
919 mutex_init(&lli->lli_och_mutex);
920 spin_lock_init(&lli->lli_agl_lock);
921 lli->lli_has_smd = false;
922 lli->lli_layout_gen = LL_LAYOUT_GEN_NONE;
923 lli->lli_clob = NULL;
925 LASSERT(lli->lli_vfs_inode.i_mode != 0);
926 if (S_ISDIR(lli->lli_vfs_inode.i_mode)) {
927 mutex_init(&lli->lli_readdir_mutex);
928 lli->lli_opendir_key = NULL;
930 lli->lli_def_acl = NULL;
931 spin_lock_init(&lli->lli_sa_lock);
932 lli->lli_opendir_pid = 0;
934 sema_init(&lli->lli_size_sem, 1);
935 lli->lli_size_sem_owner = NULL;
936 lli->lli_symlink_name = NULL;
937 init_rwsem(&lli->lli_trunc_sem);
938 mutex_init(&lli->lli_write_mutex);
939 init_rwsem(&lli->lli_glimpse_sem);
940 lli->lli_glimpse_time = 0;
941 INIT_LIST_HEAD(&lli->lli_agl_list);
942 lli->lli_agl_index = 0;
943 lli->lli_async_rc = 0;
944 lli->lli_volatile = false;
946 mutex_init(&lli->lli_layout_mutex);
949 static inline int ll_bdi_register(struct backing_dev_info *bdi)
951 static atomic_t ll_bdi_num = ATOMIC_INIT(0);
953 bdi->name = "lustre";
954 return bdi_register(bdi, NULL, "lustre-%d",
955 atomic_inc_return(&ll_bdi_num));
958 int ll_fill_super(struct super_block *sb, struct vfsmount *mnt)
960 struct lustre_profile *lprof = NULL;
961 struct lustre_sb_info *lsi = s2lsi(sb);
962 struct ll_sb_info *sbi;
963 char *dt = NULL, *md = NULL;
964 char *profilenm = get_profile_name(sb);
965 struct config_llog_instance *cfg;
966 /* %p for void* in printf needs 16+2 characters: 0xffffffffffffffff */
967 const int instlen = sizeof(cfg->cfg_instance) * 2 + 2;
970 CDEBUG(D_VFSTRACE, "VFS Op: sb %p\n", sb);
976 try_module_get(THIS_MODULE);
978 /* client additional sb info */
979 lsi->lsi_llsbi = sbi = ll_init_sbi();
981 module_put(THIS_MODULE);
986 err = ll_options(lsi->lsi_lmd->lmd_opts, &sbi->ll_flags);
990 err = bdi_init(&lsi->lsi_bdi);
993 lsi->lsi_flags |= LSI_BDI_INITIALIZED;
994 lsi->lsi_bdi.capabilities = BDI_CAP_MAP_COPY;
995 err = ll_bdi_register(&lsi->lsi_bdi);
999 sb->s_bdi = &lsi->lsi_bdi;
1001 /* Generate a string unique to this super, in case some joker tries
1002 to mount the same fs at two mount points.
1003 Use the address of the super itself.*/
1004 cfg->cfg_instance = sb;
1005 cfg->cfg_uuid = lsi->lsi_llsbi->ll_sb_uuid;
1006 cfg->cfg_callback = class_config_llog_handler;
1007 /* set up client obds */
1008 err = lustre_process_log(sb, profilenm, cfg);
1010 CERROR("Unable to process log: %d\n", err);
1011 GOTO(out_free, err);
1014 /* Profile set with LCFG_MOUNTOPT so we can find our mdc and osc obds */
1015 lprof = class_get_profile(profilenm);
1016 if (lprof == NULL) {
1017 LCONSOLE_ERROR_MSG(0x156, "The client profile '%s' could not be"
1018 " read from the MGS. Does that filesystem "
1019 "exist?\n", profilenm);
1020 GOTO(out_free, err = -EINVAL);
1022 CDEBUG(D_CONFIG, "Found profile %s: mdc=%s osc=%s\n", profilenm,
1023 lprof->lp_md, lprof->lp_dt);
1025 OBD_ALLOC(dt, strlen(lprof->lp_dt) + instlen + 2);
1027 GOTO(out_free, err = -ENOMEM);
1028 sprintf(dt, "%s-%p", lprof->lp_dt, cfg->cfg_instance);
1030 OBD_ALLOC(md, strlen(lprof->lp_md) + instlen + 2);
1032 GOTO(out_free, err = -ENOMEM);
1033 sprintf(md, "%s-%p", lprof->lp_md, cfg->cfg_instance);
1035 /* connections, registrations, sb setup */
1036 err = client_common_fill_super(sb, md, dt, mnt);
1040 OBD_FREE(md, strlen(lprof->lp_md) + instlen + 2);
1042 OBD_FREE(dt, strlen(lprof->lp_dt) + instlen + 2);
1045 else if (sbi->ll_flags & LL_SBI_VERBOSE)
1046 LCONSOLE_WARN("Mounted %s\n", profilenm);
1050 } /* ll_fill_super */
1052 void ll_put_super(struct super_block *sb)
1054 struct config_llog_instance cfg;
1055 struct obd_device *obd;
1056 struct lustre_sb_info *lsi = s2lsi(sb);
1057 struct ll_sb_info *sbi = ll_s2sbi(sb);
1058 char *profilenm = get_profile_name(sb);
1059 int next, force = 1;
1061 CDEBUG(D_VFSTRACE, "VFS Op: sb %p - %s\n", sb, profilenm);
1063 ll_print_capa_stat(sbi);
1065 cfg.cfg_instance = sb;
1066 lustre_end_log(sb, profilenm, &cfg);
1068 if (sbi->ll_md_exp) {
1069 obd = class_exp2obd(sbi->ll_md_exp);
1071 force = obd->obd_force;
1074 /* We need to set force before the lov_disconnect in
1075 lustre_common_put_super, since l_d cleans up osc's as well. */
1078 while ((obd = class_devices_in_group(&sbi->ll_sb_uuid,
1080 obd->obd_force = force;
1085 /* Only if client_common_fill_super succeeded */
1086 client_common_put_super(sb);
1090 while ((obd = class_devices_in_group(&sbi->ll_sb_uuid, &next)) !=NULL) {
1091 class_manual_cleanup(obd);
1094 if (sbi->ll_flags & LL_SBI_VERBOSE)
1095 LCONSOLE_WARN("Unmounted %s\n", profilenm ? profilenm : "");
1098 class_del_profile(profilenm);
1100 if (lsi->lsi_flags & LSI_BDI_INITIALIZED) {
1101 bdi_destroy(&lsi->lsi_bdi);
1102 lsi->lsi_flags &= ~LSI_BDI_INITIALIZED;
1106 lsi->lsi_llsbi = NULL;
1108 lustre_common_put_super(sb);
1110 module_put(THIS_MODULE);
1111 } /* client_put_super */
1113 struct inode *ll_inode_from_resource_lock(struct ldlm_lock *lock)
1115 struct inode *inode = NULL;
1117 /* NOTE: we depend on atomic igrab() -bzzz */
1118 lock_res_and_lock(lock);
1119 if (lock->l_resource->lr_lvb_inode) {
1120 struct ll_inode_info * lli;
1121 lli = ll_i2info(lock->l_resource->lr_lvb_inode);
1122 if (lli->lli_inode_magic == LLI_INODE_MAGIC) {
1123 inode = igrab(lock->l_resource->lr_lvb_inode);
1125 inode = lock->l_resource->lr_lvb_inode;
1126 LDLM_DEBUG_LIMIT(inode->i_state & I_FREEING ? D_INFO :
1127 D_WARNING, lock, "lr_lvb_inode %p is "
1128 "bogus: magic %08x",
1129 lock->l_resource->lr_lvb_inode,
1130 lli->lli_inode_magic);
1134 unlock_res_and_lock(lock);
1138 struct inode *ll_inode_from_lock(struct ldlm_lock *lock)
1140 struct inode *inode = NULL;
1141 /* NOTE: we depend on atomic igrab() -bzzz */
1142 lock_res_and_lock(lock);
1143 if (lock->l_ast_data) {
1144 struct ll_inode_info *lli = ll_i2info(lock->l_ast_data);
1145 if (lli->lli_inode_magic == LLI_INODE_MAGIC) {
1146 inode = igrab(lock->l_ast_data);
1148 inode = lock->l_ast_data;
1149 LDLM_DEBUG_LIMIT(inode->i_state & I_FREEING ? D_INFO :
1150 D_WARNING, lock, "l_ast_data %p is "
1151 "bogus: magic %08x", lock->l_ast_data,
1152 lli->lli_inode_magic);
1156 unlock_res_and_lock(lock);
1160 void ll_clear_inode(struct inode *inode)
1162 struct ll_inode_info *lli = ll_i2info(inode);
1163 struct ll_sb_info *sbi = ll_i2sbi(inode);
1165 CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p)\n", inode->i_ino,
1166 inode->i_generation, inode);
1168 if (S_ISDIR(inode->i_mode)) {
1169 /* these should have been cleared in ll_file_release */
1170 LASSERT(lli->lli_opendir_key == NULL);
1171 LASSERT(lli->lli_sai == NULL);
1172 LASSERT(lli->lli_opendir_pid == 0);
1175 spin_lock(&lli->lli_lock);
1176 ll_i2info(inode)->lli_flags &= ~LLIF_MDS_SIZE_LOCK;
1177 spin_unlock(&lli->lli_lock);
1178 md_null_inode(sbi->ll_md_exp, ll_inode2fid(inode));
1180 LASSERT(!lli->lli_open_fd_write_count);
1181 LASSERT(!lli->lli_open_fd_read_count);
1182 LASSERT(!lli->lli_open_fd_exec_count);
1184 if (lli->lli_mds_write_och)
1185 ll_md_real_close(inode, FMODE_WRITE);
1186 if (lli->lli_mds_exec_och)
1187 ll_md_real_close(inode, FMODE_EXEC);
1188 if (lli->lli_mds_read_och)
1189 ll_md_real_close(inode, FMODE_READ);
1191 if (S_ISLNK(inode->i_mode) && lli->lli_symlink_name) {
1192 OBD_FREE(lli->lli_symlink_name,
1193 strlen(lli->lli_symlink_name) + 1);
1194 lli->lli_symlink_name = NULL;
1197 if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
1198 LASSERT(lli->lli_posix_acl == NULL);
1199 if (lli->lli_remote_perms) {
1200 free_rmtperm_hash(lli->lli_remote_perms);
1201 lli->lli_remote_perms = NULL;
1204 #ifdef CONFIG_FS_POSIX_ACL
1205 else if (lli->lli_posix_acl) {
1206 LASSERT(atomic_read(&lli->lli_posix_acl->a_refcount) == 1);
1207 LASSERT(lli->lli_remote_perms == NULL);
1208 posix_acl_release(lli->lli_posix_acl);
1209 lli->lli_posix_acl = NULL;
1212 lli->lli_inode_magic = LLI_INODE_DEAD;
1214 ll_clear_inode_capas(inode);
1215 if (!S_ISDIR(inode->i_mode))
1216 LASSERT(list_empty(&lli->lli_agl_list));
1219 * XXX This has to be done before lsm is freed below, because
1220 * cl_object still uses inode lsm.
1222 cl_inode_fini(inode);
1223 lli->lli_has_smd = false;
1226 int ll_md_setattr(struct dentry *dentry, struct md_op_data *op_data,
1227 struct md_open_data **mod)
1229 struct lustre_md md;
1230 struct inode *inode = dentry->d_inode;
1231 struct ll_sb_info *sbi = ll_i2sbi(inode);
1232 struct ptlrpc_request *request = NULL;
1235 op_data = ll_prep_md_op_data(op_data, inode, NULL, NULL, 0, 0,
1236 LUSTRE_OPC_ANY, NULL);
1237 if (IS_ERR(op_data))
1238 return PTR_ERR(op_data);
1240 rc = md_setattr(sbi->ll_md_exp, op_data, NULL, 0, NULL, 0,
1243 ptlrpc_req_finished(request);
1244 if (rc == -ENOENT) {
1246 /* Unlinked special device node? Or just a race?
1247 * Pretend we done everything. */
1248 if (!S_ISREG(inode->i_mode) &&
1249 !S_ISDIR(inode->i_mode)) {
1250 ia_valid = op_data->op_attr.ia_valid;
1251 op_data->op_attr.ia_valid &= ~TIMES_SET_FLAGS;
1252 rc = simple_setattr(dentry, &op_data->op_attr);
1253 op_data->op_attr.ia_valid = ia_valid;
1255 } else if (rc != -EPERM && rc != -EACCES && rc != -ETXTBSY) {
1256 CERROR("md_setattr fails: rc = %d\n", rc);
1261 rc = md_get_lustre_md(sbi->ll_md_exp, request, sbi->ll_dt_exp,
1262 sbi->ll_md_exp, &md);
1264 ptlrpc_req_finished(request);
1268 ia_valid = op_data->op_attr.ia_valid;
1269 /* inode size will be in ll_setattr_ost, can't do it now since dirty
1270 * cache is not cleared yet. */
1271 op_data->op_attr.ia_valid &= ~(TIMES_SET_FLAGS | ATTR_SIZE);
1272 rc = simple_setattr(dentry, &op_data->op_attr);
1273 op_data->op_attr.ia_valid = ia_valid;
1275 /* Extract epoch data if obtained. */
1276 op_data->op_handle = md.body->handle;
1277 op_data->op_ioepoch = md.body->ioepoch;
1279 ll_update_inode(inode, &md);
1280 ptlrpc_req_finished(request);
1285 /* Close IO epoch and send Size-on-MDS attribute update. */
1286 static int ll_setattr_done_writing(struct inode *inode,
1287 struct md_op_data *op_data,
1288 struct md_open_data *mod)
1290 struct ll_inode_info *lli = ll_i2info(inode);
1293 LASSERT(op_data != NULL);
1294 if (!S_ISREG(inode->i_mode))
1297 CDEBUG(D_INODE, "Epoch "LPU64" closed on "DFID" for truncate\n",
1298 op_data->op_ioepoch, PFID(&lli->lli_fid));
1300 op_data->op_flags = MF_EPOCH_CLOSE;
1301 ll_done_writing_attr(inode, op_data);
1302 ll_pack_inode2opdata(inode, op_data, NULL);
1304 rc = md_done_writing(ll_i2sbi(inode)->ll_md_exp, op_data, mod);
1305 if (rc == -EAGAIN) {
1306 /* MDS has instructed us to obtain Size-on-MDS attribute
1307 * from OSTs and send setattr to back to MDS. */
1308 rc = ll_som_update(inode, op_data);
1310 CERROR("inode %lu mdc truncate failed: rc = %d\n",
1316 static int ll_setattr_ost(struct inode *inode, struct iattr *attr)
1318 struct obd_capa *capa;
1321 if (attr->ia_valid & ATTR_SIZE)
1322 capa = ll_osscapa_get(inode, CAPA_OPC_OSS_TRUNC);
1324 capa = ll_mdscapa_get(inode);
1326 rc = cl_setattr_ost(inode, attr, capa);
1328 if (attr->ia_valid & ATTR_SIZE)
1329 ll_truncate_free_capa(capa);
1337 /* If this inode has objects allocated to it (lsm != NULL), then the OST
1338 * object(s) determine the file size and mtime. Otherwise, the MDS will
1339 * keep these values until such a time that objects are allocated for it.
1340 * We do the MDS operations first, as it is checking permissions for us.
1341 * We don't to the MDS RPC if there is nothing that we want to store there,
1342 * otherwise there is no harm in updating mtime/atime on the MDS if we are
1343 * going to do an RPC anyways.
1345 * If we are doing a truncate, we will send the mtime and ctime updates
1346 * to the OST with the punch RPC, otherwise we do an explicit setattr RPC.
1347 * I don't believe it is possible to get e.g. ATTR_MTIME_SET and ATTR_SIZE
1350 int ll_setattr_raw(struct dentry *dentry, struct iattr *attr)
1352 struct inode *inode = dentry->d_inode;
1353 struct ll_inode_info *lli = ll_i2info(inode);
1354 struct md_op_data *op_data = NULL;
1355 struct md_open_data *mod = NULL;
1356 int rc = 0, rc1 = 0;
1358 CDEBUG(D_VFSTRACE, "%s: setattr inode %p/fid:"DFID" from %llu to %llu, "
1359 "valid %x\n", ll_get_fsname(inode->i_sb, NULL, 0), inode,
1360 PFID(&lli->lli_fid), i_size_read(inode), attr->ia_size,
1363 if (attr->ia_valid & ATTR_SIZE) {
1364 /* Check new size against VFS/VM file size limit and rlimit */
1365 rc = inode_newsize_ok(inode, attr->ia_size);
1369 /* The maximum Lustre file size is variable, based on the
1370 * OST maximum object size and number of stripes. This
1371 * needs another check in addition to the VFS check above. */
1372 if (attr->ia_size > ll_file_maxbytes(inode)) {
1373 CDEBUG(D_INODE,"file "DFID" too large %llu > "LPU64"\n",
1374 PFID(&lli->lli_fid), attr->ia_size,
1375 ll_file_maxbytes(inode));
1379 attr->ia_valid |= ATTR_MTIME | ATTR_CTIME;
1382 /* POSIX: check before ATTR_*TIME_SET set (from inode_change_ok) */
1383 if (attr->ia_valid & TIMES_SET_FLAGS) {
1384 if ((!uid_eq(current_fsuid(), inode->i_uid)) &&
1385 !cfs_capable(CFS_CAP_FOWNER))
1389 /* We mark all of the fields "set" so MDS/OST does not re-set them */
1390 if (attr->ia_valid & ATTR_CTIME) {
1391 attr->ia_ctime = CURRENT_TIME;
1392 attr->ia_valid |= ATTR_CTIME_SET;
1394 if (!(attr->ia_valid & ATTR_ATIME_SET) &&
1395 (attr->ia_valid & ATTR_ATIME)) {
1396 attr->ia_atime = CURRENT_TIME;
1397 attr->ia_valid |= ATTR_ATIME_SET;
1399 if (!(attr->ia_valid & ATTR_MTIME_SET) &&
1400 (attr->ia_valid & ATTR_MTIME)) {
1401 attr->ia_mtime = CURRENT_TIME;
1402 attr->ia_valid |= ATTR_MTIME_SET;
1405 if (attr->ia_valid & (ATTR_MTIME | ATTR_CTIME))
1406 CDEBUG(D_INODE, "setting mtime %lu, ctime %lu, now = %lu\n",
1407 LTIME_S(attr->ia_mtime), LTIME_S(attr->ia_ctime),
1408 cfs_time_current_sec());
1410 /* If we are changing file size, file content is modified, flag it. */
1411 if (attr->ia_valid & ATTR_SIZE) {
1412 attr->ia_valid |= MDS_OPEN_OWNEROVERRIDE;
1413 spin_lock(&lli->lli_lock);
1414 lli->lli_flags |= LLIF_DATA_MODIFIED;
1415 spin_unlock(&lli->lli_lock);
1418 /* We always do an MDS RPC, even if we're only changing the size;
1419 * only the MDS knows whether truncate() should fail with -ETXTBUSY */
1421 OBD_ALLOC_PTR(op_data);
1422 if (op_data == NULL)
1425 if (!S_ISDIR(inode->i_mode)) {
1426 if (attr->ia_valid & ATTR_SIZE)
1427 inode_dio_write_done(inode);
1428 mutex_unlock(&inode->i_mutex);
1429 down_write(&lli->lli_trunc_sem);
1432 memcpy(&op_data->op_attr, attr, sizeof(*attr));
1434 /* Open epoch for truncate. */
1435 if (exp_connect_som(ll_i2mdexp(inode)) &&
1436 (attr->ia_valid & (ATTR_SIZE | ATTR_MTIME | ATTR_MTIME_SET)))
1437 op_data->op_flags = MF_EPOCH_OPEN;
1439 rc = ll_md_setattr(dentry, op_data, &mod);
1443 /* RPC to MDT is sent, cancel data modification flag */
1444 if (rc == 0 && (op_data->op_bias & MDS_DATA_MODIFIED)) {
1445 spin_lock(&lli->lli_lock);
1446 lli->lli_flags &= ~LLIF_DATA_MODIFIED;
1447 spin_unlock(&lli->lli_lock);
1450 ll_ioepoch_open(lli, op_data->op_ioepoch);
1451 if (!S_ISREG(inode->i_mode))
1454 if (attr->ia_valid & (ATTR_SIZE |
1455 ATTR_ATIME | ATTR_ATIME_SET |
1456 ATTR_MTIME | ATTR_MTIME_SET))
1457 /* For truncate and utimes sending attributes to OSTs, setting
1458 * mtime/atime to the past will be performed under PW [0:EOF]
1459 * extent lock (new_size:EOF for truncate). It may seem
1460 * excessive to send mtime/atime updates to OSTs when not
1461 * setting times to past, but it is necessary due to possible
1462 * time de-synchronization between MDT inode and OST objects */
1463 rc = ll_setattr_ost(inode, attr);
1466 if (op_data->op_ioepoch) {
1467 rc1 = ll_setattr_done_writing(inode, op_data, mod);
1471 ll_finish_md_op_data(op_data);
1473 if (!S_ISDIR(inode->i_mode)) {
1474 up_write(&lli->lli_trunc_sem);
1475 mutex_lock(&inode->i_mutex);
1476 if (attr->ia_valid & ATTR_SIZE)
1477 inode_dio_wait(inode);
1480 ll_stats_ops_tally(ll_i2sbi(inode), (attr->ia_valid & ATTR_SIZE) ?
1481 LPROC_LL_TRUNC : LPROC_LL_SETATTR, 1);
1486 int ll_setattr(struct dentry *de, struct iattr *attr)
1488 int mode = de->d_inode->i_mode;
1490 if ((attr->ia_valid & (ATTR_CTIME|ATTR_SIZE|ATTR_MODE)) ==
1491 (ATTR_CTIME|ATTR_SIZE|ATTR_MODE))
1492 attr->ia_valid |= MDS_OPEN_OWNEROVERRIDE;
1494 if (((attr->ia_valid & (ATTR_MODE|ATTR_FORCE|ATTR_SIZE)) ==
1495 (ATTR_SIZE|ATTR_MODE)) &&
1496 (((mode & S_ISUID) && !(attr->ia_mode & S_ISUID)) ||
1497 (((mode & (S_ISGID|S_IXGRP)) == (S_ISGID|S_IXGRP)) &&
1498 !(attr->ia_mode & S_ISGID))))
1499 attr->ia_valid |= ATTR_FORCE;
1501 if ((mode & S_ISUID) &&
1502 !(attr->ia_mode & S_ISUID) &&
1503 !(attr->ia_valid & ATTR_KILL_SUID))
1504 attr->ia_valid |= ATTR_KILL_SUID;
1506 if (((mode & (S_ISGID|S_IXGRP)) == (S_ISGID|S_IXGRP)) &&
1507 !(attr->ia_mode & S_ISGID) &&
1508 !(attr->ia_valid & ATTR_KILL_SGID))
1509 attr->ia_valid |= ATTR_KILL_SGID;
1511 return ll_setattr_raw(de, attr);
1514 int ll_statfs_internal(struct super_block *sb, struct obd_statfs *osfs,
1515 __u64 max_age, __u32 flags)
1517 struct ll_sb_info *sbi = ll_s2sbi(sb);
1518 struct obd_statfs obd_osfs;
1521 rc = obd_statfs(NULL, sbi->ll_md_exp, osfs, max_age, flags);
1523 CERROR("md_statfs fails: rc = %d\n", rc);
1527 osfs->os_type = sb->s_magic;
1529 CDEBUG(D_SUPER, "MDC blocks "LPU64"/"LPU64" objects "LPU64"/"LPU64"\n",
1530 osfs->os_bavail, osfs->os_blocks, osfs->os_ffree,osfs->os_files);
1532 if (sbi->ll_flags & LL_SBI_LAZYSTATFS)
1533 flags |= OBD_STATFS_NODELAY;
1535 rc = obd_statfs_rqset(sbi->ll_dt_exp, &obd_osfs, max_age, flags);
1537 CERROR("obd_statfs fails: rc = %d\n", rc);
1541 CDEBUG(D_SUPER, "OSC blocks "LPU64"/"LPU64" objects "LPU64"/"LPU64"\n",
1542 obd_osfs.os_bavail, obd_osfs.os_blocks, obd_osfs.os_ffree,
1545 osfs->os_bsize = obd_osfs.os_bsize;
1546 osfs->os_blocks = obd_osfs.os_blocks;
1547 osfs->os_bfree = obd_osfs.os_bfree;
1548 osfs->os_bavail = obd_osfs.os_bavail;
1550 /* If we don't have as many objects free on the OST as inodes
1551 * on the MDS, we reduce the total number of inodes to
1552 * compensate, so that the "inodes in use" number is correct.
1554 if (obd_osfs.os_ffree < osfs->os_ffree) {
1555 osfs->os_files = (osfs->os_files - osfs->os_ffree) +
1557 osfs->os_ffree = obd_osfs.os_ffree;
1562 int ll_statfs(struct dentry *de, struct kstatfs *sfs)
1564 struct super_block *sb = de->d_sb;
1565 struct obd_statfs osfs;
1568 CDEBUG(D_VFSTRACE, "VFS Op: at "LPU64" jiffies\n", get_jiffies_64());
1569 ll_stats_ops_tally(ll_s2sbi(sb), LPROC_LL_STAFS, 1);
1571 /* Some amount of caching on the client is allowed */
1572 rc = ll_statfs_internal(sb, &osfs,
1573 cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS),
1578 statfs_unpack(sfs, &osfs);
1580 /* We need to downshift for all 32-bit kernels, because we can't
1581 * tell if the kernel is being called via sys_statfs64() or not.
1582 * Stop before overflowing f_bsize - in which case it is better
1583 * to just risk EOVERFLOW if caller is using old sys_statfs(). */
1584 if (sizeof(long) < 8) {
1585 while (osfs.os_blocks > ~0UL && sfs->f_bsize < 0x40000000) {
1588 osfs.os_blocks >>= 1;
1589 osfs.os_bfree >>= 1;
1590 osfs.os_bavail >>= 1;
1594 sfs->f_blocks = osfs.os_blocks;
1595 sfs->f_bfree = osfs.os_bfree;
1596 sfs->f_bavail = osfs.os_bavail;
1597 sfs->f_fsid = ll_s2sbi(sb)->ll_fsid;
1601 void ll_inode_size_lock(struct inode *inode)
1603 struct ll_inode_info *lli;
1605 LASSERT(!S_ISDIR(inode->i_mode));
1607 lli = ll_i2info(inode);
1608 LASSERT(lli->lli_size_sem_owner != current);
1609 down(&lli->lli_size_sem);
1610 LASSERT(lli->lli_size_sem_owner == NULL);
1611 lli->lli_size_sem_owner = current;
1614 void ll_inode_size_unlock(struct inode *inode)
1616 struct ll_inode_info *lli;
1618 lli = ll_i2info(inode);
1619 LASSERT(lli->lli_size_sem_owner == current);
1620 lli->lli_size_sem_owner = NULL;
1621 up(&lli->lli_size_sem);
1624 void ll_update_inode(struct inode *inode, struct lustre_md *md)
1626 struct ll_inode_info *lli = ll_i2info(inode);
1627 struct mdt_body *body = md->body;
1628 struct lov_stripe_md *lsm = md->lsm;
1629 struct ll_sb_info *sbi = ll_i2sbi(inode);
1631 LASSERT ((lsm != NULL) == ((body->valid & OBD_MD_FLEASIZE) != 0));
1633 if (!lli->lli_has_smd &&
1634 !(sbi->ll_flags & LL_SBI_LAYOUT_LOCK))
1635 cl_file_inode_init(inode, md);
1637 lli->lli_maxbytes = lsm->lsm_maxbytes;
1638 if (lli->lli_maxbytes > MAX_LFS_FILESIZE)
1639 lli->lli_maxbytes = MAX_LFS_FILESIZE;
1642 if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
1643 if (body->valid & OBD_MD_FLRMTPERM)
1644 ll_update_remote_perm(inode, md->remote_perm);
1646 #ifdef CONFIG_FS_POSIX_ACL
1647 else if (body->valid & OBD_MD_FLACL) {
1648 spin_lock(&lli->lli_lock);
1649 if (lli->lli_posix_acl)
1650 posix_acl_release(lli->lli_posix_acl);
1651 lli->lli_posix_acl = md->posix_acl;
1652 spin_unlock(&lli->lli_lock);
1655 inode->i_ino = cl_fid_build_ino(&body->fid1,
1656 sbi->ll_flags & LL_SBI_32BIT_API);
1657 inode->i_generation = cl_fid_build_gen(&body->fid1);
1659 if (body->valid & OBD_MD_FLATIME) {
1660 if (body->atime > LTIME_S(inode->i_atime))
1661 LTIME_S(inode->i_atime) = body->atime;
1662 lli->lli_lvb.lvb_atime = body->atime;
1664 if (body->valid & OBD_MD_FLMTIME) {
1665 if (body->mtime > LTIME_S(inode->i_mtime)) {
1666 CDEBUG(D_INODE, "setting ino %lu mtime from %lu "
1667 "to "LPU64"\n", inode->i_ino,
1668 LTIME_S(inode->i_mtime), body->mtime);
1669 LTIME_S(inode->i_mtime) = body->mtime;
1671 lli->lli_lvb.lvb_mtime = body->mtime;
1673 if (body->valid & OBD_MD_FLCTIME) {
1674 if (body->ctime > LTIME_S(inode->i_ctime))
1675 LTIME_S(inode->i_ctime) = body->ctime;
1676 lli->lli_lvb.lvb_ctime = body->ctime;
1678 if (body->valid & OBD_MD_FLMODE)
1679 inode->i_mode = (inode->i_mode & S_IFMT)|(body->mode & ~S_IFMT);
1680 if (body->valid & OBD_MD_FLTYPE)
1681 inode->i_mode = (inode->i_mode & ~S_IFMT)|(body->mode & S_IFMT);
1682 LASSERT(inode->i_mode != 0);
1683 if (S_ISREG(inode->i_mode)) {
1684 inode->i_blkbits = min(PTLRPC_MAX_BRW_BITS + 1, LL_MAX_BLKSIZE_BITS);
1686 inode->i_blkbits = inode->i_sb->s_blocksize_bits;
1688 if (body->valid & OBD_MD_FLUID)
1689 inode->i_uid = make_kuid(&init_user_ns, body->uid);
1690 if (body->valid & OBD_MD_FLGID)
1691 inode->i_gid = make_kgid(&init_user_ns, body->gid);
1692 if (body->valid & OBD_MD_FLFLAGS)
1693 inode->i_flags = ll_ext_to_inode_flags(body->flags);
1694 if (body->valid & OBD_MD_FLNLINK)
1695 set_nlink(inode, body->nlink);
1696 if (body->valid & OBD_MD_FLRDEV)
1697 inode->i_rdev = old_decode_dev(body->rdev);
1699 if (body->valid & OBD_MD_FLID) {
1700 /* FID shouldn't be changed! */
1701 if (fid_is_sane(&lli->lli_fid)) {
1702 LASSERTF(lu_fid_eq(&lli->lli_fid, &body->fid1),
1703 "Trying to change FID "DFID
1704 " to the "DFID", inode %lu/%u(%p)\n",
1705 PFID(&lli->lli_fid), PFID(&body->fid1),
1706 inode->i_ino, inode->i_generation, inode);
1708 lli->lli_fid = body->fid1;
1711 LASSERT(fid_seq(&lli->lli_fid) != 0);
1713 if (body->valid & OBD_MD_FLSIZE) {
1714 if (exp_connect_som(ll_i2mdexp(inode)) &&
1715 S_ISREG(inode->i_mode)) {
1716 struct lustre_handle lockh;
1719 /* As it is possible a blocking ast has been processed
1720 * by this time, we need to check there is an UPDATE
1721 * lock on the client and set LLIF_MDS_SIZE_LOCK holding
1723 mode = ll_take_md_lock(inode, MDS_INODELOCK_UPDATE,
1724 &lockh, LDLM_FL_CBPENDING);
1726 if (lli->lli_flags & (LLIF_DONE_WRITING |
1727 LLIF_EPOCH_PENDING |
1729 CERROR("ino %lu flags %u still has "
1730 "size authority! do not trust "
1731 "the size got from MDS\n",
1732 inode->i_ino, lli->lli_flags);
1734 /* Use old size assignment to avoid
1735 * deadlock bz14138 & bz14326 */
1736 i_size_write(inode, body->size);
1737 spin_lock(&lli->lli_lock);
1738 lli->lli_flags |= LLIF_MDS_SIZE_LOCK;
1739 spin_unlock(&lli->lli_lock);
1741 ldlm_lock_decref(&lockh, mode);
1744 /* Use old size assignment to avoid
1745 * deadlock bz14138 & bz14326 */
1746 i_size_write(inode, body->size);
1748 CDEBUG(D_VFSTRACE, "inode=%lu, updating i_size %llu\n",
1749 inode->i_ino, (unsigned long long)body->size);
1752 if (body->valid & OBD_MD_FLBLOCKS)
1753 inode->i_blocks = body->blocks;
1756 if (body->valid & OBD_MD_FLMDSCAPA) {
1757 LASSERT(md->mds_capa);
1758 ll_add_capa(inode, md->mds_capa);
1760 if (body->valid & OBD_MD_FLOSSCAPA) {
1761 LASSERT(md->oss_capa);
1762 ll_add_capa(inode, md->oss_capa);
1766 void ll_read_inode2(struct inode *inode, void *opaque)
1768 struct lustre_md *md = opaque;
1769 struct ll_inode_info *lli = ll_i2info(inode);
1771 CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p)\n",
1772 PFID(&lli->lli_fid), inode);
1774 LASSERT(!lli->lli_has_smd);
1776 /* Core attributes from the MDS first. This is a new inode, and
1777 * the VFS doesn't zero times in the core inode so we have to do
1778 * it ourselves. They will be overwritten by either MDS or OST
1779 * attributes - we just need to make sure they aren't newer. */
1780 LTIME_S(inode->i_mtime) = 0;
1781 LTIME_S(inode->i_atime) = 0;
1782 LTIME_S(inode->i_ctime) = 0;
1784 ll_update_inode(inode, md);
1786 /* OIDEBUG(inode); */
1788 /* initializing backing dev info. */
1789 inode->i_mapping->backing_dev_info = &s2lsi(inode->i_sb)->lsi_bdi;
1792 if (S_ISREG(inode->i_mode)) {
1793 struct ll_sb_info *sbi = ll_i2sbi(inode);
1794 inode->i_op = &ll_file_inode_operations;
1795 inode->i_fop = sbi->ll_fop;
1796 inode->i_mapping->a_ops = (struct address_space_operations *)&ll_aops;
1797 } else if (S_ISDIR(inode->i_mode)) {
1798 inode->i_op = &ll_dir_inode_operations;
1799 inode->i_fop = &ll_dir_operations;
1800 } else if (S_ISLNK(inode->i_mode)) {
1801 inode->i_op = &ll_fast_symlink_inode_operations;
1803 inode->i_op = &ll_special_inode_operations;
1805 init_special_inode(inode, inode->i_mode,
1810 void ll_delete_inode(struct inode *inode)
1812 struct cl_inode_info *lli = cl_i2info(inode);
1814 if (S_ISREG(inode->i_mode) && lli->lli_clob != NULL)
1815 /* discard all dirty pages before truncating them, required by
1816 * osc_extent implementation at LU-1030. */
1817 cl_sync_file_range(inode, 0, OBD_OBJECT_EOF,
1818 CL_FSYNC_DISCARD, 1);
1820 truncate_inode_pages(&inode->i_data, 0);
1822 /* Workaround for LU-118 */
1823 if (inode->i_data.nrpages) {
1824 TREE_READ_LOCK_IRQ(&inode->i_data);
1825 TREE_READ_UNLOCK_IRQ(&inode->i_data);
1826 LASSERTF(inode->i_data.nrpages == 0,
1827 "inode=%lu/%u(%p) nrpages=%lu, see "
1828 "http://jira.whamcloud.com/browse/LU-118\n",
1829 inode->i_ino, inode->i_generation, inode,
1830 inode->i_data.nrpages);
1832 /* Workaround end */
1834 ll_clear_inode(inode);
1838 int ll_iocontrol(struct inode *inode, struct file *file,
1839 unsigned int cmd, unsigned long arg)
1841 struct ll_sb_info *sbi = ll_i2sbi(inode);
1842 struct ptlrpc_request *req = NULL;
1846 case FSFILT_IOC_GETFLAGS: {
1847 struct mdt_body *body;
1848 struct md_op_data *op_data;
1850 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL,
1851 0, 0, LUSTRE_OPC_ANY,
1853 if (IS_ERR(op_data))
1854 return PTR_ERR(op_data);
1856 op_data->op_valid = OBD_MD_FLFLAGS;
1857 rc = md_getattr(sbi->ll_md_exp, op_data, &req);
1858 ll_finish_md_op_data(op_data);
1860 CERROR("failure %d inode %lu\n", rc, inode->i_ino);
1864 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
1866 flags = body->flags;
1868 ptlrpc_req_finished(req);
1870 return put_user(flags, (int *)arg);
1872 case FSFILT_IOC_SETFLAGS: {
1873 struct lov_stripe_md *lsm;
1874 struct obd_info oinfo = { { { 0 } } };
1875 struct md_op_data *op_data;
1877 if (get_user(flags, (int *)arg))
1880 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
1881 LUSTRE_OPC_ANY, NULL);
1882 if (IS_ERR(op_data))
1883 return PTR_ERR(op_data);
1885 ((struct ll_iattr *)&op_data->op_attr)->ia_attr_flags = flags;
1886 op_data->op_attr.ia_valid |= ATTR_ATTR_FLAG;
1887 rc = md_setattr(sbi->ll_md_exp, op_data,
1888 NULL, 0, NULL, 0, &req, NULL);
1889 ll_finish_md_op_data(op_data);
1890 ptlrpc_req_finished(req);
1894 inode->i_flags = ll_ext_to_inode_flags(flags);
1896 lsm = ccc_inode_lsm_get(inode);
1897 if (!lsm_has_objects(lsm)) {
1898 ccc_inode_lsm_put(inode, lsm);
1902 OBDO_ALLOC(oinfo.oi_oa);
1904 ccc_inode_lsm_put(inode, lsm);
1908 oinfo.oi_oa->o_oi = lsm->lsm_oi;
1909 oinfo.oi_oa->o_flags = flags;
1910 oinfo.oi_oa->o_valid = OBD_MD_FLID | OBD_MD_FLFLAGS |
1912 oinfo.oi_capa = ll_mdscapa_get(inode);
1913 obdo_set_parent_fid(oinfo.oi_oa, &ll_i2info(inode)->lli_fid);
1914 rc = obd_setattr_rqset(sbi->ll_dt_exp, &oinfo, NULL);
1915 capa_put(oinfo.oi_capa);
1916 OBDO_FREE(oinfo.oi_oa);
1917 ccc_inode_lsm_put(inode, lsm);
1919 if (rc && rc != -EPERM && rc != -EACCES)
1920 CERROR("osc_setattr_async fails: rc = %d\n", rc);
1931 int ll_flush_ctx(struct inode *inode)
1933 struct ll_sb_info *sbi = ll_i2sbi(inode);
1935 CDEBUG(D_SEC, "flush context for user %d\n",
1936 from_kuid(&init_user_ns, current_uid()));
1938 obd_set_info_async(NULL, sbi->ll_md_exp,
1939 sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
1941 obd_set_info_async(NULL, sbi->ll_dt_exp,
1942 sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
1947 /* umount -f client means force down, don't save state */
1948 void ll_umount_begin(struct super_block *sb)
1950 struct ll_sb_info *sbi = ll_s2sbi(sb);
1951 struct obd_device *obd;
1952 struct obd_ioctl_data *ioc_data;
1954 CDEBUG(D_VFSTRACE, "VFS Op: superblock %p count %d active %d\n", sb,
1955 sb->s_count, atomic_read(&sb->s_active));
1957 obd = class_exp2obd(sbi->ll_md_exp);
1959 CERROR("Invalid MDC connection handle "LPX64"\n",
1960 sbi->ll_md_exp->exp_handle.h_cookie);
1965 obd = class_exp2obd(sbi->ll_dt_exp);
1967 CERROR("Invalid LOV connection handle "LPX64"\n",
1968 sbi->ll_dt_exp->exp_handle.h_cookie);
1973 OBD_ALLOC_PTR(ioc_data);
1975 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_md_exp,
1976 sizeof(*ioc_data), ioc_data, NULL);
1978 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_dt_exp,
1979 sizeof(*ioc_data), ioc_data, NULL);
1981 OBD_FREE_PTR(ioc_data);
1984 /* Really, we'd like to wait until there are no requests outstanding,
1985 * and then continue. For now, we just invalidate the requests,
1986 * schedule() and sleep one second if needed, and hope.
1991 int ll_remount_fs(struct super_block *sb, int *flags, char *data)
1993 struct ll_sb_info *sbi = ll_s2sbi(sb);
1994 char *profilenm = get_profile_name(sb);
1998 if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY)) {
1999 read_only = *flags & MS_RDONLY;
2000 err = obd_set_info_async(NULL, sbi->ll_md_exp,
2001 sizeof(KEY_READ_ONLY),
2002 KEY_READ_ONLY, sizeof(read_only),
2005 LCONSOLE_WARN("Failed to remount %s %s (%d)\n",
2006 profilenm, read_only ?
2007 "read-only" : "read-write", err);
2012 sb->s_flags |= MS_RDONLY;
2014 sb->s_flags &= ~MS_RDONLY;
2016 if (sbi->ll_flags & LL_SBI_VERBOSE)
2017 LCONSOLE_WARN("Remounted %s %s\n", profilenm,
2018 read_only ? "read-only" : "read-write");
2023 int ll_prep_inode(struct inode **inode, struct ptlrpc_request *req,
2024 struct super_block *sb, struct lookup_intent *it)
2026 struct ll_sb_info *sbi = NULL;
2027 struct lustre_md md;
2030 LASSERT(*inode || sb);
2031 sbi = sb ? ll_s2sbi(sb) : ll_i2sbi(*inode);
2032 rc = md_get_lustre_md(sbi->ll_md_exp, req, sbi->ll_dt_exp,
2033 sbi->ll_md_exp, &md);
2038 ll_update_inode(*inode, &md);
2040 LASSERT(sb != NULL);
2043 * At this point server returns to client's same fid as client
2044 * generated for creating. So using ->fid1 is okay here.
2046 LASSERT(fid_is_sane(&md.body->fid1));
2048 *inode = ll_iget(sb, cl_fid_build_ino(&md.body->fid1,
2049 sbi->ll_flags & LL_SBI_32BIT_API),
2051 if (*inode == NULL || IS_ERR(*inode)) {
2052 #ifdef CONFIG_FS_POSIX_ACL
2054 posix_acl_release(md.posix_acl);
2055 md.posix_acl = NULL;
2058 rc = IS_ERR(*inode) ? PTR_ERR(*inode) : -ENOMEM;
2060 CERROR("new_inode -fatal: rc %d\n", rc);
2065 /* Handling piggyback layout lock.
2066 * Layout lock can be piggybacked by getattr and open request.
2067 * The lsm can be applied to inode only if it comes with a layout lock
2068 * otherwise correct layout may be overwritten, for example:
2069 * 1. proc1: mdt returns a lsm but not granting layout
2070 * 2. layout was changed by another client
2071 * 3. proc2: refresh layout and layout lock granted
2072 * 4. proc1: to apply a stale layout */
2073 if (it != NULL && it->d.lustre.it_lock_mode != 0) {
2074 struct lustre_handle lockh;
2075 struct ldlm_lock *lock;
2077 lockh.cookie = it->d.lustre.it_lock_handle;
2078 lock = ldlm_handle2lock(&lockh);
2079 LASSERT(lock != NULL);
2080 if (ldlm_has_layout(lock)) {
2081 struct cl_object_conf conf;
2083 memset(&conf, 0, sizeof(conf));
2084 conf.coc_opc = OBJECT_CONF_SET;
2085 conf.coc_inode = *inode;
2086 conf.coc_lock = lock;
2087 conf.u.coc_md = &md;
2088 (void)ll_layout_conf(*inode, &conf);
2090 LDLM_LOCK_PUT(lock);
2095 obd_free_memmd(sbi->ll_dt_exp, &md.lsm);
2096 md_free_lustre_md(sbi->ll_md_exp, &md);
2100 int ll_obd_statfs(struct inode *inode, void *arg)
2102 struct ll_sb_info *sbi = NULL;
2103 struct obd_export *exp;
2105 struct obd_ioctl_data *data = NULL;
2110 if (!inode || !(sbi = ll_i2sbi(inode)))
2111 GOTO(out_statfs, rc = -EINVAL);
2113 rc = obd_ioctl_getdata(&buf, &len, arg);
2115 GOTO(out_statfs, rc);
2118 if (!data->ioc_inlbuf1 || !data->ioc_inlbuf2 ||
2119 !data->ioc_pbuf1 || !data->ioc_pbuf2)
2120 GOTO(out_statfs, rc = -EINVAL);
2122 if (data->ioc_inllen1 != sizeof(__u32) ||
2123 data->ioc_inllen2 != sizeof(__u32) ||
2124 data->ioc_plen1 != sizeof(struct obd_statfs) ||
2125 data->ioc_plen2 != sizeof(struct obd_uuid))
2126 GOTO(out_statfs, rc = -EINVAL);
2128 memcpy(&type, data->ioc_inlbuf1, sizeof(__u32));
2129 if (type & LL_STATFS_LMV)
2130 exp = sbi->ll_md_exp;
2131 else if (type & LL_STATFS_LOV)
2132 exp = sbi->ll_dt_exp;
2134 GOTO(out_statfs, rc = -ENODEV);
2136 flags = (type & LL_STATFS_NODELAY) ? OBD_STATFS_NODELAY : 0;
2137 rc = obd_iocontrol(IOC_OBD_STATFS, exp, len, buf, &flags);
2139 GOTO(out_statfs, rc);
2142 obd_ioctl_freedata(buf, len);
2146 int ll_process_config(struct lustre_cfg *lcfg)
2150 struct lprocfs_static_vars lvars;
2154 lprocfs_llite_init_vars(&lvars);
2156 /* The instance name contains the sb: lustre-client-aacfe000 */
2157 ptr = strrchr(lustre_cfg_string(lcfg, 0), '-');
2158 if (!ptr || !*(++ptr))
2160 if (sscanf(ptr, "%lx", &x) != 1)
2163 /* This better be a real Lustre superblock! */
2164 LASSERT(s2lsi((struct super_block *)sb)->lsi_lmd->lmd_magic == LMD_MAGIC);
2166 /* Note we have not called client_common_fill_super yet, so
2167 proc fns must be able to handle that! */
2168 rc = class_process_proc_param(PARAM_LLITE, lvars.obd_vars,
2175 /* this function prepares md_op_data hint for passing ot down to MD stack. */
2176 struct md_op_data * ll_prep_md_op_data(struct md_op_data *op_data,
2177 struct inode *i1, struct inode *i2,
2178 const char *name, int namelen,
2179 int mode, __u32 opc, void *data)
2181 LASSERT(i1 != NULL);
2183 if (namelen > ll_i2sbi(i1)->ll_namelen)
2184 return ERR_PTR(-ENAMETOOLONG);
2186 if (op_data == NULL)
2187 OBD_ALLOC_PTR(op_data);
2189 if (op_data == NULL)
2190 return ERR_PTR(-ENOMEM);
2192 ll_i2gids(op_data->op_suppgids, i1, i2);
2193 op_data->op_fid1 = *ll_inode2fid(i1);
2194 op_data->op_capa1 = ll_mdscapa_get(i1);
2197 op_data->op_fid2 = *ll_inode2fid(i2);
2198 op_data->op_capa2 = ll_mdscapa_get(i2);
2200 fid_zero(&op_data->op_fid2);
2201 op_data->op_capa2 = NULL;
2204 op_data->op_name = name;
2205 op_data->op_namelen = namelen;
2206 op_data->op_mode = mode;
2207 op_data->op_mod_time = cfs_time_current_sec();
2208 op_data->op_fsuid = from_kuid(&init_user_ns, current_fsuid());
2209 op_data->op_fsgid = from_kgid(&init_user_ns, current_fsgid());
2210 op_data->op_cap = cfs_curproc_cap_pack();
2211 op_data->op_bias = 0;
2212 op_data->op_cli_flags = 0;
2213 if ((opc == LUSTRE_OPC_CREATE) && (name != NULL) &&
2214 filename_is_volatile(name, namelen, NULL))
2215 op_data->op_bias |= MDS_CREATE_VOLATILE;
2216 op_data->op_opc = opc;
2217 op_data->op_mds = 0;
2218 op_data->op_data = data;
2220 /* If the file is being opened after mknod() (normally due to NFS)
2221 * try to use the default stripe data from parent directory for
2222 * allocating OST objects. Try to pass the parent FID to MDS. */
2223 if (opc == LUSTRE_OPC_CREATE && i1 == i2 && S_ISREG(i2->i_mode) &&
2224 !ll_i2info(i2)->lli_has_smd) {
2225 struct ll_inode_info *lli = ll_i2info(i2);
2227 spin_lock(&lli->lli_lock);
2228 if (likely(!lli->lli_has_smd && !fid_is_zero(&lli->lli_pfid)))
2229 op_data->op_fid1 = lli->lli_pfid;
2230 spin_unlock(&lli->lli_lock);
2231 /** We ignore parent's capability temporary. */
2234 /* When called by ll_setattr_raw, file is i1. */
2235 if (LLIF_DATA_MODIFIED & ll_i2info(i1)->lli_flags)
2236 op_data->op_bias |= MDS_DATA_MODIFIED;
2241 void ll_finish_md_op_data(struct md_op_data *op_data)
2243 capa_put(op_data->op_capa1);
2244 capa_put(op_data->op_capa2);
2245 OBD_FREE_PTR(op_data);
2248 int ll_show_options(struct seq_file *seq, struct dentry *dentry)
2250 struct ll_sb_info *sbi;
2252 LASSERT((seq != NULL) && (dentry != NULL));
2253 sbi = ll_s2sbi(dentry->d_sb);
2255 if (sbi->ll_flags & LL_SBI_NOLCK)
2256 seq_puts(seq, ",nolock");
2258 if (sbi->ll_flags & LL_SBI_FLOCK)
2259 seq_puts(seq, ",flock");
2261 if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
2262 seq_puts(seq, ",localflock");
2264 if (sbi->ll_flags & LL_SBI_USER_XATTR)
2265 seq_puts(seq, ",user_xattr");
2267 if (sbi->ll_flags & LL_SBI_LAZYSTATFS)
2268 seq_puts(seq, ",lazystatfs");
2270 if (sbi->ll_flags & LL_SBI_USER_FID2PATH)
2271 seq_puts(seq, ",user_fid2path");
2277 * Get obd name by cmd, and copy out to user space
2279 int ll_get_obd_name(struct inode *inode, unsigned int cmd, unsigned long arg)
2281 struct ll_sb_info *sbi = ll_i2sbi(inode);
2282 struct obd_device *obd;
2284 if (cmd == OBD_IOC_GETDTNAME)
2285 obd = class_exp2obd(sbi->ll_dt_exp);
2286 else if (cmd == OBD_IOC_GETMDNAME)
2287 obd = class_exp2obd(sbi->ll_md_exp);
2294 if (copy_to_user((void *)arg, obd->obd_name,
2295 strlen(obd->obd_name) + 1))
2302 * Get lustre file system name by \a sbi. If \a buf is provided(non-NULL), the
2303 * fsname will be returned in this buffer; otherwise, a static buffer will be
2304 * used to store the fsname and returned to caller.
2306 char *ll_get_fsname(struct super_block *sb, char *buf, int buflen)
2308 static char fsname_static[MTI_NAME_MAXLEN];
2309 struct lustre_sb_info *lsi = s2lsi(sb);
2314 /* this means the caller wants to use static buffer
2315 * and it doesn't care about race. Usually this is
2316 * in error reporting path */
2317 buf = fsname_static;
2318 buflen = sizeof(fsname_static);
2321 len = strlen(lsi->lsi_lmd->lmd_profile);
2322 ptr = strrchr(lsi->lsi_lmd->lmd_profile, '-');
2323 if (ptr && (strcmp(ptr, "-client") == 0))
2326 if (unlikely(len >= buflen))
2328 strncpy(buf, lsi->lsi_lmd->lmd_profile, len);
2334 static char* ll_d_path(struct dentry *dentry, char *buf, int bufsize)
2341 p.mnt = current->fs->root.mnt;
2343 path = d_path(&p, buf, bufsize);
2349 void ll_dirty_page_discard_warn(struct page *page, int ioret)
2351 char *buf, *path = NULL;
2352 struct dentry *dentry = NULL;
2353 struct ccc_object *obj = cl_inode2ccc(page->mapping->host);
2355 /* this can be called inside spin lock so use GFP_ATOMIC. */
2356 buf = (char *)__get_free_page(GFP_ATOMIC);
2358 dentry = d_find_alias(page->mapping->host);
2360 path = ll_d_path(dentry, buf, PAGE_SIZE);
2363 CWARN("%s: dirty page discard: %s/fid: "DFID"/%s may get corrupted "
2364 "(rc %d)\n", ll_get_fsname(page->mapping->host->i_sb, NULL, 0),
2365 s2lsi(page->mapping->host->i_sb)->lsi_lmd->lmd_dev,
2366 PFID(&obj->cob_header.coh_lu.loh_fid),
2367 (path && !IS_ERR(path)) ? path : "", ioret);
2373 free_page((unsigned long)buf);