1 /* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
2 * vim:expandtab:shiftwidth=8:tabstop=8:
6 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
8 * This program is free software; you can redistribute it and/or modify
9 * it under the terms of the GNU General Public License version 2 only,
10 * as published by the Free Software Foundation.
12 * This program is distributed in the hope that it will be useful, but
13 * WITHOUT ANY WARRANTY; without even the implied warranty of
14 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
15 * General Public License version 2 for more details (a copy is included
16 * in the LICENSE file that accompanied this code).
18 * You should have received a copy of the GNU General Public License
19 * version 2 along with this program; If not, see
20 * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
22 * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
23 * CA 95054 USA or visit www.sun.com if you need additional information or
29 * Copyright 2008 Sun Microsystems, Inc. All rights reserved
30 * Use is subject to license terms.
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>
45 #include <linux/random.h>
46 #include <linux/version.h>
49 #include <lustre_lite.h>
50 #include <lustre_ha.h>
51 #include <lustre_dlm.h>
52 #include <lprocfs_status.h>
53 #include <lustre_disk.h>
54 #include <lustre_param.h>
55 #include <lustre_log.h>
56 #include <cl_object.h>
57 #include <obd_cksum.h>
58 #include "llite_internal.h"
60 cfs_mem_cache_t *ll_file_data_slab;
62 CFS_LIST_HEAD(ll_super_blocks);
63 cfs_spinlock_t ll_sb_lock = CFS_SPIN_LOCK_UNLOCKED;
65 extern struct address_space_operations ll_aops;
66 extern struct address_space_operations ll_dir_aops;
69 #define log2(n) cfs_ffz(~(n))
72 static struct ll_sb_info *ll_init_sbi(void)
74 struct ll_sb_info *sbi = NULL;
81 OBD_ALLOC(sbi, sizeof(*sbi));
85 cfs_spin_lock_init(&sbi->ll_lock);
86 cfs_init_mutex(&sbi->ll_lco.lco_lock);
87 cfs_spin_lock_init(&sbi->ll_pp_extent_lock);
88 cfs_spin_lock_init(&sbi->ll_process_lock);
89 sbi->ll_rw_stats_on = 0;
92 pages = si.totalram - si.totalhigh;
93 if (pages >> (20 - CFS_PAGE_SHIFT) < 512) {
94 #ifdef HAVE_BGL_SUPPORT
95 sbi->ll_async_page_max = pages / 4;
97 sbi->ll_async_page_max = pages / 2;
100 sbi->ll_async_page_max = (pages / 4) * 3;
103 sbi->ll_ra_info.ra_max_pages_per_file = min(pages / 32,
104 SBI_DEFAULT_READAHEAD_MAX);
105 sbi->ll_ra_info.ra_max_pages = sbi->ll_ra_info.ra_max_pages_per_file;
106 sbi->ll_ra_info.ra_max_read_ahead_whole_pages =
107 SBI_DEFAULT_READAHEAD_WHOLE_MAX;
108 CFS_INIT_LIST_HEAD(&sbi->ll_conn_chain);
109 CFS_INIT_LIST_HEAD(&sbi->ll_orphan_dentry_list);
111 ll_generate_random_uuid(uuid);
112 class_uuid_unparse(uuid, &sbi->ll_sb_uuid);
113 CDEBUG(D_CONFIG, "generated uuid: %s\n", sbi->ll_sb_uuid.uuid);
115 cfs_spin_lock(&ll_sb_lock);
116 cfs_list_add_tail(&sbi->ll_list, &ll_super_blocks);
117 cfs_spin_unlock(&ll_sb_lock);
119 #ifdef ENABLE_LLITE_CHECKSUM
120 sbi->ll_flags |= LL_SBI_CHECKSUM;
123 #ifdef HAVE_LRU_RESIZE_SUPPORT
124 sbi->ll_flags |= LL_SBI_LRU_RESIZE;
127 for (i = 0; i <= LL_PROCESS_HIST_MAX; i++) {
128 cfs_spin_lock_init(&sbi->ll_rw_extents_info.pp_extents[i]. \
130 cfs_spin_lock_init(&sbi->ll_rw_extents_info.pp_extents[i]. \
134 /* metadata statahead is enabled by default */
135 sbi->ll_sa_max = LL_SA_RPC_DEF;
136 atomic_set(&sbi->ll_sa_total, 0);
137 atomic_set(&sbi->ll_sa_wrong, 0);
142 void ll_free_sbi(struct super_block *sb)
144 struct ll_sb_info *sbi = ll_s2sbi(sb);
148 cfs_spin_lock(&ll_sb_lock);
149 cfs_list_del(&sbi->ll_list);
150 cfs_spin_unlock(&ll_sb_lock);
151 OBD_FREE(sbi, sizeof(*sbi));
156 static struct dentry_operations ll_d_root_ops = {
157 .d_compare = ll_dcompare,
158 .d_revalidate = ll_revalidate_nd,
161 static int client_common_fill_super(struct super_block *sb, char *md, char *dt)
163 struct inode *root = 0;
164 struct ll_sb_info *sbi = ll_s2sbi(sb);
165 struct obd_device *obd;
166 struct obd_capa *oc = NULL;
167 struct obd_statfs osfs;
168 struct ptlrpc_request *request = NULL;
169 struct obd_connect_data *data = NULL;
170 struct obd_uuid *uuid;
171 struct md_op_data *op_data;
172 struct lustre_md lmd;
174 int size, err, checksum;
177 obd = class_name2obd(md);
179 CERROR("MD %s: not setup or attached\n", md);
187 if (proc_lustre_fs_root) {
188 err = lprocfs_register_mountpoint(proc_lustre_fs_root, sb,
191 CERROR("could not register mount in /proc/fs/lustre\n");
194 /* indicate the features supported by this client */
195 data->ocd_connect_flags = OBD_CONNECT_IBITS | OBD_CONNECT_NODEVOH |
196 OBD_CONNECT_JOIN | OBD_CONNECT_ATTRFID |
197 OBD_CONNECT_VERSION | OBD_CONNECT_MDS_CAPA |
198 OBD_CONNECT_OSS_CAPA | OBD_CONNECT_CANCELSET|
199 OBD_CONNECT_FID | OBD_CONNECT_AT |
200 OBD_CONNECT_LOV_V3 | OBD_CONNECT_RMT_CLIENT |
201 OBD_CONNECT_VBR | OBD_CONNECT_SOM;
203 #ifdef HAVE_LRU_RESIZE_SUPPORT
204 if (sbi->ll_flags & LL_SBI_LRU_RESIZE)
205 data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
207 #ifdef CONFIG_FS_POSIX_ACL
208 data->ocd_connect_flags |= OBD_CONNECT_ACL;
210 data->ocd_ibits_known = MDS_INODELOCK_FULL;
211 data->ocd_version = LUSTRE_VERSION_CODE;
213 if (sb->s_flags & MS_RDONLY)
214 data->ocd_connect_flags |= OBD_CONNECT_RDONLY;
215 if (sbi->ll_flags & LL_SBI_USER_XATTR)
216 data->ocd_connect_flags |= OBD_CONNECT_XATTR;
218 #ifdef HAVE_MS_FLOCK_LOCK
219 /* force vfs to use lustre handler for flock() calls - bug 10743 */
220 sb->s_flags |= MS_FLOCK_LOCK;
223 if (sbi->ll_flags & LL_SBI_FLOCK)
224 sbi->ll_fop = &ll_file_operations_flock;
225 else if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
226 sbi->ll_fop = &ll_file_operations;
228 sbi->ll_fop = &ll_file_operations_noflock;
231 data->ocd_connect_flags |= OBD_CONNECT_REAL;
232 if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
233 data->ocd_connect_flags |= OBD_CONNECT_RMT_CLIENT_FORCE;
235 err = obd_connect(NULL, &sbi->ll_md_exp, obd, &sbi->ll_sb_uuid, data, NULL);
237 LCONSOLE_ERROR_MSG(0x14f, "An MDT (md %s) is performing "
238 "recovery, of which this client is not a "
239 "part. Please wait for recovery to complete,"
240 " abort, or time out.\n", md);
243 CERROR("cannot connect to %s: rc = %d\n", md, err);
247 err = obd_fid_init(sbi->ll_md_exp);
249 CERROR("Can't init metadata layer FID infrastructure, "
254 err = obd_statfs(obd, &osfs, cfs_time_current_64() - CFS_HZ, 0);
256 GOTO(out_md_fid, err);
258 size = sizeof(*data);
259 err = obd_get_info(sbi->ll_md_exp, sizeof(KEY_CONN_DATA),
260 KEY_CONN_DATA, &size, data, NULL);
262 CERROR("Get connect data failed: %d \n", err);
266 LASSERT(osfs.os_bsize);
267 sb->s_blocksize = osfs.os_bsize;
268 sb->s_blocksize_bits = log2(osfs.os_bsize);
269 sb->s_magic = LL_SUPER_MAGIC;
271 /* for bug 11559. in $LINUX/fs/read_write.c, function do_sendfile():
272 * retval = in_file->f_op->sendfile(...);
274 * retval = -EOVERFLOW;
276 * it will check if *ppos is greater than max. However, max equals to
277 * s_maxbytes, which is a negative integer in a x86_64 box since loff_t
278 * has been defined as a signed long long integer in linux kernel. */
279 #if BITS_PER_LONG == 64
280 sb->s_maxbytes = PAGE_CACHE_MAXBYTES >> 1;
282 sb->s_maxbytes = PAGE_CACHE_MAXBYTES;
284 sbi->ll_namelen = osfs.os_namelen;
285 sbi->ll_max_rw_chunk = LL_DEFAULT_MAX_RW_CHUNK;
287 if ((sbi->ll_flags & LL_SBI_USER_XATTR) &&
288 !(data->ocd_connect_flags & OBD_CONNECT_XATTR)) {
289 LCONSOLE_INFO("Disabling user_xattr feature because "
290 "it is not supported on the server\n");
291 sbi->ll_flags &= ~LL_SBI_USER_XATTR;
294 if (data->ocd_connect_flags & OBD_CONNECT_ACL) {
296 sb->s_flags |= MS_POSIXACL;
298 sbi->ll_flags |= LL_SBI_ACL;
300 LCONSOLE_INFO("client wants to enable acl, but mdt not!\n");
302 sb->s_flags &= ~MS_POSIXACL;
304 sbi->ll_flags &= ~LL_SBI_ACL;
307 if (data->ocd_connect_flags & OBD_CONNECT_RMT_CLIENT) {
308 if (!(sbi->ll_flags & LL_SBI_RMT_CLIENT)) {
309 sbi->ll_flags |= LL_SBI_RMT_CLIENT;
310 LCONSOLE_INFO("client is set as remote by default.\n");
313 if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
314 sbi->ll_flags &= ~LL_SBI_RMT_CLIENT;
315 LCONSOLE_INFO("client claims to be remote, but server "
316 "rejected, forced to be local.\n");
320 if (data->ocd_connect_flags & OBD_CONNECT_MDS_CAPA) {
321 LCONSOLE_INFO("client enabled MDS capability!\n");
322 sbi->ll_flags |= LL_SBI_MDS_CAPA;
325 if (data->ocd_connect_flags & OBD_CONNECT_OSS_CAPA) {
326 LCONSOLE_INFO("client enabled OSS capability!\n");
327 sbi->ll_flags |= LL_SBI_OSS_CAPA;
330 obd = class_name2obd(dt);
332 CERROR("DT %s: not setup or attached\n", dt);
333 GOTO(out_md_fid, err = -ENODEV);
336 data->ocd_connect_flags = OBD_CONNECT_GRANT | OBD_CONNECT_VERSION |
337 OBD_CONNECT_REQPORTAL | OBD_CONNECT_BRW_SIZE |
338 OBD_CONNECT_CANCELSET | OBD_CONNECT_FID |
339 OBD_CONNECT_SRVLOCK | OBD_CONNECT_TRUNCLOCK|
340 OBD_CONNECT_AT | OBD_CONNECT_RMT_CLIENT |
341 OBD_CONNECT_OSS_CAPA | OBD_CONNECT_VBR|
344 if (!OBD_FAIL_CHECK(OBD_FAIL_OSC_CONNECT_CKSUM)) {
345 /* OBD_CONNECT_CKSUM should always be set, even if checksums are
346 * disabled by default, because it can still be enabled on the
347 * fly via /proc. As a consequence, we still need to come to an
348 * agreement on the supported algorithms at connect time */
349 data->ocd_connect_flags |= OBD_CONNECT_CKSUM;
351 if (OBD_FAIL_CHECK(OBD_FAIL_OSC_CKSUM_ADLER_ONLY))
352 data->ocd_cksum_types = OBD_CKSUM_ADLER;
354 /* send the list of supported checksum types */
355 data->ocd_cksum_types = OBD_CKSUM_ALL;
358 #ifdef HAVE_LRU_RESIZE_SUPPORT
359 data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
361 if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
362 data->ocd_connect_flags |= OBD_CONNECT_RMT_CLIENT_FORCE;
364 CDEBUG(D_RPCTRACE, "ocd_connect_flags: "LPX64" ocd_version: %d "
365 "ocd_grant: %d\n", data->ocd_connect_flags,
366 data->ocd_version, data->ocd_grant);
368 obd->obd_upcall.onu_owner = &sbi->ll_lco;
369 obd->obd_upcall.onu_upcall = cl_ocd_update;
370 data->ocd_brw_size = PTLRPC_MAX_BRW_PAGES << CFS_PAGE_SHIFT;
372 err = obd_connect(NULL, &sbi->ll_dt_exp, obd, &sbi->ll_sb_uuid, data, NULL);
374 LCONSOLE_ERROR_MSG(0x150, "An OST (dt %s) is performing "
375 "recovery, of which this client is not a "
376 "part. Please wait for recovery to "
377 "complete, abort, or time out.\n", dt);
378 GOTO(out_md_fid, err);
380 CERROR("Cannot connect to %s: rc = %d\n", dt, err);
381 GOTO(out_md_fid, err);
384 err = obd_fid_init(sbi->ll_dt_exp);
386 CERROR("Can't init data layer FID infrastructure, "
391 cfs_mutex_down(&sbi->ll_lco.lco_lock);
392 sbi->ll_lco.lco_flags = data->ocd_connect_flags;
393 sbi->ll_lco.lco_md_exp = sbi->ll_md_exp;
394 sbi->ll_lco.lco_dt_exp = sbi->ll_dt_exp;
395 cfs_mutex_up(&sbi->ll_lco.lco_lock);
397 fid_zero(&sbi->ll_root_fid);
398 err = md_getstatus(sbi->ll_md_exp, &sbi->ll_root_fid, &oc);
400 CERROR("cannot mds_connect: rc = %d\n", err);
401 GOTO(out_lock_cn_cb, err);
403 if (!fid_is_sane(&sbi->ll_root_fid)) {
404 CERROR("Invalid root fid during mount\n");
405 GOTO(out_lock_cn_cb, err = -EINVAL);
407 CDEBUG(D_SUPER, "rootfid "DFID"\n", PFID(&sbi->ll_root_fid));
409 sb->s_op = &lustre_super_operations;
410 sb->s_export_op = &lustre_export_operations;
413 * XXX: move this to after cbd setup? */
414 valid = OBD_MD_FLGETATTR | OBD_MD_FLBLOCKS | OBD_MD_FLMDSCAPA;
415 if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
416 valid |= OBD_MD_FLRMTPERM;
417 else if (sbi->ll_flags & LL_SBI_ACL)
418 valid |= OBD_MD_FLACL;
420 OBD_ALLOC_PTR(op_data);
422 GOTO(out_lock_cn_cb, err = -ENOMEM);
424 op_data->op_fid1 = sbi->ll_root_fid;
425 op_data->op_mode = 0;
426 op_data->op_capa1 = oc;
427 op_data->op_valid = valid;
429 err = md_getattr(sbi->ll_md_exp, op_data, &request);
432 OBD_FREE_PTR(op_data);
434 CERROR("md_getattr failed for root: rc = %d\n", err);
435 GOTO(out_lock_cn_cb, err);
437 memset(&lmd, 0, sizeof(lmd));
438 err = md_get_lustre_md(sbi->ll_md_exp, request, sbi->ll_dt_exp,
439 sbi->ll_md_exp, &lmd);
441 CERROR("failed to understand root inode md: rc = %d\n", err);
442 ptlrpc_req_finished (request);
443 GOTO(out_lock_cn_cb, err);
446 LASSERT(fid_is_sane(&sbi->ll_root_fid));
447 root = ll_iget(sb, cl_fid_build_ino(&sbi->ll_root_fid), &lmd);
448 md_free_lustre_md(sbi->ll_md_exp, &lmd);
449 ptlrpc_req_finished(request);
451 if (root == NULL || IS_ERR(root)) {
453 obd_free_memmd(sbi->ll_dt_exp, &lmd.lsm);
454 #ifdef CONFIG_FS_POSIX_ACL
456 posix_acl_release(lmd.posix_acl);
457 lmd.posix_acl = NULL;
460 err = IS_ERR(root) ? PTR_ERR(root) : -EBADF;
462 CERROR("lustre_lite: bad iget4 for root\n");
466 err = ll_close_thread_start(&sbi->ll_lcq);
468 CERROR("cannot start close thread: rc %d\n", err);
472 #ifdef CONFIG_FS_POSIX_ACL
473 if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
474 rct_init(&sbi->ll_rct);
475 et_init(&sbi->ll_et);
479 checksum = sbi->ll_flags & LL_SBI_CHECKSUM;
480 err = obd_set_info_async(sbi->ll_dt_exp, sizeof(KEY_CHECKSUM),
481 KEY_CHECKSUM, sizeof(checksum), &checksum,
485 sb->s_root = d_alloc_root(root);
487 OBD_FREE(data, sizeof(*data));
489 sb->s_root->d_op = &ll_d_root_ops;
491 sbi->ll_sdev_orig = sb->s_dev;
493 /* We set sb->s_dev equal on all lustre clients in order to support
494 * NFS export clustering. NFSD requires that the FSID be the same
496 /* s_dev is also used in lt_compare() to compare two fs, but that is
497 * only a node-local comparison. */
498 uuid = obd_get_uuid(sbi->ll_md_exp);
500 sb->s_dev = get_uuid2int(uuid->uuid, strlen(uuid->uuid));
507 obd_fid_fini(sbi->ll_dt_exp);
509 obd_disconnect(sbi->ll_dt_exp);
510 sbi->ll_dt_exp = NULL;
512 obd_fid_fini(sbi->ll_md_exp);
514 obd_disconnect(sbi->ll_md_exp);
515 sbi->ll_md_exp = NULL;
519 lprocfs_unregister_mountpoint(sbi);
523 int ll_get_max_mdsize(struct ll_sb_info *sbi, int *lmmsize)
527 *lmmsize = obd_size_diskmd(sbi->ll_dt_exp, NULL);
529 rc = obd_get_info(sbi->ll_md_exp, sizeof(KEY_MAX_EASIZE),
530 KEY_MAX_EASIZE, &size, lmmsize, NULL);
532 CERROR("Get max mdsize error rc %d \n", rc);
537 void ll_dump_inode(struct inode *inode)
539 struct list_head *tmp;
540 int dentry_count = 0;
542 LASSERT(inode != NULL);
544 list_for_each(tmp, &inode->i_dentry)
547 CERROR("inode %p dump: dev=%s ino=%lu mode=%o count=%u, %d dentries\n",
548 inode, ll_i2mdexp(inode)->exp_obd->obd_name, inode->i_ino,
549 inode->i_mode, atomic_read(&inode->i_count), dentry_count);
552 void lustre_dump_dentry(struct dentry *dentry, int recur)
554 struct list_head *tmp;
557 LASSERT(dentry != NULL);
559 list_for_each(tmp, &dentry->d_subdirs)
562 CERROR("dentry %p dump: name=%.*s parent=%.*s (%p), inode=%p, count=%u,"
563 " flags=0x%x, fsdata=%p, %d subdirs\n", dentry,
564 dentry->d_name.len, dentry->d_name.name,
565 dentry->d_parent->d_name.len, dentry->d_parent->d_name.name,
566 dentry->d_parent, dentry->d_inode, atomic_read(&dentry->d_count),
567 dentry->d_flags, dentry->d_fsdata, subdirs);
568 if (dentry->d_inode != NULL)
569 ll_dump_inode(dentry->d_inode);
574 list_for_each(tmp, &dentry->d_subdirs) {
575 struct dentry *d = list_entry(tmp, struct dentry, d_child);
576 lustre_dump_dentry(d, recur - 1);
580 void client_common_put_super(struct super_block *sb)
582 struct ll_sb_info *sbi = ll_s2sbi(sb);
585 #ifdef CONFIG_FS_POSIX_ACL
586 if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
587 et_fini(&sbi->ll_et);
588 rct_fini(&sbi->ll_rct);
592 obd_cancel_unused(sbi->ll_dt_exp, NULL, 0, NULL);
594 ll_close_thread_shutdown(sbi->ll_lcq);
598 cfs_list_del(&sbi->ll_conn_chain);
600 obd_fid_fini(sbi->ll_dt_exp);
601 obd_disconnect(sbi->ll_dt_exp);
602 sbi->ll_dt_exp = NULL;
604 lprocfs_unregister_mountpoint(sbi);
606 obd_fid_fini(sbi->ll_md_exp);
607 obd_disconnect(sbi->ll_md_exp);
608 sbi->ll_md_exp = NULL;
613 void ll_kill_super(struct super_block *sb)
615 struct ll_sb_info *sbi;
620 if (!(sb->s_flags & MS_ACTIVE))
624 /* we need restore s_dev from changed for clustred NFS before put_super
625 * because new kernels have cached s_dev and change sb->s_dev in
626 * put_super not affected real removing devices */
628 sb->s_dev = sbi->ll_sdev_orig;
632 char *ll_read_opt(const char *opt, char *data)
638 CDEBUG(D_SUPER, "option: %s, data %s\n", opt, data);
639 if (strncmp(opt, data, strlen(opt)))
641 if ((value = strchr(data, '=')) == NULL)
645 OBD_ALLOC(retval, strlen(value) + 1);
647 CERROR("out of memory!\n");
651 memcpy(retval, value, strlen(value)+1);
652 CDEBUG(D_SUPER, "Assigned option: %s, value %s\n", opt, retval);
656 static inline int ll_set_opt(const char *opt, char *data, int fl)
658 if (strncmp(opt, data, strlen(opt)) != 0)
664 /* non-client-specific mount options are parsed in lmd_parse */
665 static int ll_options(char *options, int *flags)
668 char *s1 = options, *s2;
674 CDEBUG(D_CONFIG, "Parsing opts %s\n", options);
677 CDEBUG(D_SUPER, "next opt=%s\n", s1);
678 tmp = ll_set_opt("nolock", s1, LL_SBI_NOLCK);
683 tmp = ll_set_opt("flock", s1, LL_SBI_FLOCK);
688 tmp = ll_set_opt("localflock", s1, LL_SBI_LOCALFLOCK);
693 tmp = ll_set_opt("noflock", s1, LL_SBI_FLOCK|LL_SBI_LOCALFLOCK);
698 tmp = ll_set_opt("user_xattr", s1, LL_SBI_USER_XATTR);
703 tmp = ll_set_opt("nouser_xattr", s1, LL_SBI_USER_XATTR);
708 tmp = ll_set_opt("acl", s1, LL_SBI_ACL);
710 /* Ignore deprecated mount option. The client will
711 * always try to mount with ACL support, whether this
712 * is used depends on whether server supports it. */
715 tmp = ll_set_opt("noacl", s1, LL_SBI_ACL);
719 tmp = ll_set_opt("remote_client", s1, LL_SBI_RMT_CLIENT);
725 tmp = ll_set_opt("checksum", s1, LL_SBI_CHECKSUM);
730 tmp = ll_set_opt("nochecksum", s1, LL_SBI_CHECKSUM);
735 tmp = ll_set_opt("lruresize", s1, LL_SBI_LRU_RESIZE);
740 tmp = ll_set_opt("nolruresize", s1, LL_SBI_LRU_RESIZE);
745 tmp = ll_set_opt("lazystatfs", s1, LL_SBI_LAZYSTATFS);
750 tmp = ll_set_opt("nolazystatfs", s1, LL_SBI_LAZYSTATFS);
756 LCONSOLE_ERROR_MSG(0x152, "Unknown option '%s', won't mount.\n",
762 s2 = strchr(s1, ',');
770 void ll_lli_init(struct ll_inode_info *lli)
772 lli->lli_inode_magic = LLI_INODE_MAGIC;
773 cfs_sema_init(&lli->lli_size_sem, 1);
774 cfs_sema_init(&lli->lli_write_sem, 1);
775 cfs_sema_init(&lli->lli_trunc_sem, 1);
777 lli->lli_maxbytes = PAGE_CACHE_MAXBYTES;
778 cfs_spin_lock_init(&lli->lli_lock);
779 CFS_INIT_LIST_HEAD(&lli->lli_close_list);
780 lli->lli_inode_magic = LLI_INODE_MAGIC;
781 cfs_sema_init(&lli->lli_och_sem, 1);
782 lli->lli_mds_read_och = lli->lli_mds_write_och = NULL;
783 lli->lli_mds_exec_och = NULL;
784 lli->lli_open_fd_read_count = lli->lli_open_fd_write_count = 0;
785 lli->lli_open_fd_exec_count = 0;
786 CFS_INIT_LIST_HEAD(&lli->lli_dead_list);
787 lli->lli_remote_perms = NULL;
788 lli->lli_rmtperm_utime = 0;
789 cfs_sema_init(&lli->lli_rmtperm_sem, 1);
790 CFS_INIT_LIST_HEAD(&lli->lli_oss_capas);
793 int ll_fill_super(struct super_block *sb)
795 struct lustre_profile *lprof;
796 struct lustre_sb_info *lsi = s2lsi(sb);
797 struct ll_sb_info *sbi;
798 char *dt = NULL, *md = NULL;
799 char *profilenm = get_profile_name(sb);
800 struct config_llog_instance cfg = {0, };
801 char ll_instance[sizeof(sb) * 2 + 1];
805 CDEBUG(D_VFSTRACE, "VFS Op: sb %p\n", sb);
809 /* client additional sb info */
810 lsi->lsi_llsbi = sbi = ll_init_sbi();
812 cfs_module_put(THIS_MODULE);
816 err = ll_options(lsi->lsi_lmd->lmd_opts, &sbi->ll_flags);
820 /* Generate a string unique to this super, in case some joker tries
821 to mount the same fs at two mount points.
822 Use the address of the super itself.*/
823 sprintf(ll_instance, "%p", sb);
824 cfg.cfg_instance = ll_instance;
825 cfg.cfg_uuid = lsi->lsi_llsbi->ll_sb_uuid;
827 /* set up client obds */
828 err = lustre_process_log(sb, profilenm, &cfg);
830 CERROR("Unable to process log: %d\n", err);
834 lprof = class_get_profile(profilenm);
836 LCONSOLE_ERROR_MSG(0x156, "The client profile '%s' could not be"
837 " read from the MGS. Does that filesystem "
838 "exist?\n", profilenm);
839 GOTO(out_free, err = -EINVAL);
841 CDEBUG(D_CONFIG, "Found profile %s: mdc=%s osc=%s\n", profilenm,
842 lprof->lp_md, lprof->lp_dt);
844 OBD_ALLOC(dt, strlen(lprof->lp_dt) +
845 strlen(ll_instance) + 2);
847 GOTO(out_free, err = -ENOMEM);
848 sprintf(dt, "%s-%s", lprof->lp_dt, ll_instance);
850 OBD_ALLOC(md, strlen(lprof->lp_md) +
851 strlen(ll_instance) + 2);
853 GOTO(out_free, err = -ENOMEM);
854 sprintf(md, "%s-%s", lprof->lp_md, ll_instance);
856 /* connections, registrations, sb setup */
857 err = client_common_fill_super(sb, md, dt);
861 OBD_FREE(md, strlen(md) + 1);
863 OBD_FREE(dt, strlen(dt) + 1);
867 LCONSOLE_WARN("Client %s has started\n", profilenm);
870 } /* ll_fill_super */
873 void lu_context_keys_dump(void);
875 void ll_put_super(struct super_block *sb)
877 struct config_llog_instance cfg;
878 char ll_instance[sizeof(sb) * 2 + 1];
879 struct obd_device *obd;
880 struct lustre_sb_info *lsi = s2lsi(sb);
881 struct ll_sb_info *sbi = ll_s2sbi(sb);
882 char *profilenm = get_profile_name(sb);
886 CDEBUG(D_VFSTRACE, "VFS Op: sb %p - %s\n", sb, profilenm);
888 ll_print_capa_stat(sbi);
890 sprintf(ll_instance, "%p", sb);
891 cfg.cfg_instance = ll_instance;
892 lustre_end_log(sb, NULL, &cfg);
894 if (sbi->ll_md_exp) {
895 obd = class_exp2obd(sbi->ll_md_exp);
897 force = obd->obd_force;
900 /* We need to set force before the lov_disconnect in
901 lustre_common_put_super, since l_d cleans up osc's as well. */
904 while ((obd = class_devices_in_group(&sbi->ll_sb_uuid,
906 obd->obd_force = force;
911 /* Only if client_common_fill_super succeeded */
912 client_common_put_super(sb);
916 while ((obd = class_devices_in_group(&sbi->ll_sb_uuid, &next)) !=NULL) {
917 class_manual_cleanup(obd);
921 class_del_profile(profilenm);
924 lsi->lsi_llsbi = NULL;
926 lustre_common_put_super(sb);
928 cl_env_cache_purge(~0);
930 LCONSOLE_WARN("client %s umount complete\n", ll_instance);
932 cfs_module_put(THIS_MODULE);
935 } /* client_put_super */
937 struct inode *ll_inode_from_lock(struct ldlm_lock *lock)
939 struct inode *inode = NULL;
940 /* NOTE: we depend on atomic igrab() -bzzz */
941 lock_res_and_lock(lock);
942 if (lock->l_ast_data) {
943 struct ll_inode_info *lli = ll_i2info(lock->l_ast_data);
944 if (lli->lli_inode_magic == LLI_INODE_MAGIC) {
945 inode = igrab(lock->l_ast_data);
947 inode = lock->l_ast_data;
948 ldlm_lock_debug(NULL, inode->i_state & I_FREEING ?
950 lock, __FILE__, __func__, __LINE__,
951 "l_ast_data %p is bogus: magic %08x",
952 lock->l_ast_data, lli->lli_inode_magic);
956 unlock_res_and_lock(lock);
960 static int null_if_equal(struct ldlm_lock *lock, void *data)
962 if (data == lock->l_ast_data) {
963 lock->l_ast_data = NULL;
965 if (lock->l_req_mode != lock->l_granted_mode)
966 LDLM_ERROR(lock,"clearing inode with ungranted lock");
969 return LDLM_ITER_CONTINUE;
972 void ll_clear_inode(struct inode *inode)
974 struct ll_inode_info *lli = ll_i2info(inode);
975 struct ll_sb_info *sbi = ll_i2sbi(inode);
978 CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p)\n", inode->i_ino,
979 inode->i_generation, inode);
981 if (S_ISDIR(inode->i_mode)) {
982 /* these should have been cleared in ll_file_release */
983 LASSERT(lli->lli_sai == NULL);
984 LASSERT(lli->lli_opendir_key == NULL);
985 LASSERT(lli->lli_opendir_pid == 0);
988 ll_i2info(inode)->lli_flags &= ~LLIF_MDS_SIZE_LOCK;
989 md_change_cbdata(sbi->ll_md_exp, ll_inode2fid(inode),
990 null_if_equal, inode);
992 LASSERT(!lli->lli_open_fd_write_count);
993 LASSERT(!lli->lli_open_fd_read_count);
994 LASSERT(!lli->lli_open_fd_exec_count);
996 if (lli->lli_mds_write_och)
997 ll_md_real_close(inode, FMODE_WRITE);
998 if (lli->lli_mds_exec_och)
999 ll_md_real_close(inode, FMODE_EXEC);
1000 if (lli->lli_mds_read_och)
1001 ll_md_real_close(inode, FMODE_READ);
1003 if (lli->lli_symlink_name) {
1004 OBD_FREE(lli->lli_symlink_name,
1005 strlen(lli->lli_symlink_name) + 1);
1006 lli->lli_symlink_name = NULL;
1009 if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
1010 LASSERT(lli->lli_posix_acl == NULL);
1011 if (lli->lli_remote_perms) {
1012 free_rmtperm_hash(lli->lli_remote_perms);
1013 lli->lli_remote_perms = NULL;
1016 #ifdef CONFIG_FS_POSIX_ACL
1017 else if (lli->lli_posix_acl) {
1018 LASSERT(cfs_atomic_read(&lli->lli_posix_acl->a_refcount) == 1);
1019 LASSERT(lli->lli_remote_perms == NULL);
1020 posix_acl_release(lli->lli_posix_acl);
1021 lli->lli_posix_acl = NULL;
1024 lli->lli_inode_magic = LLI_INODE_DEAD;
1026 ll_clear_inode_capas(inode);
1028 * XXX This has to be done before lsm is freed below, because
1029 * cl_object still uses inode lsm.
1031 cl_inode_fini(inode);
1034 obd_free_memmd(sbi->ll_dt_exp, &lli->lli_smd);
1035 lli->lli_smd = NULL;
1042 int ll_md_setattr(struct inode *inode, struct md_op_data *op_data,
1043 struct md_open_data **mod)
1045 struct lustre_md md;
1046 struct ll_sb_info *sbi = ll_i2sbi(inode);
1047 struct ptlrpc_request *request = NULL;
1051 op_data = ll_prep_md_op_data(op_data, inode, NULL, NULL, 0, 0,
1052 LUSTRE_OPC_ANY, NULL);
1053 if (IS_ERR(op_data))
1054 RETURN(PTR_ERR(op_data));
1056 rc = md_setattr(sbi->ll_md_exp, op_data, NULL, 0, NULL, 0,
1059 ptlrpc_req_finished(request);
1060 if (rc == -ENOENT) {
1062 /* Unlinked special device node? Or just a race?
1063 * Pretend we done everything. */
1064 if (!S_ISREG(inode->i_mode) &&
1065 !S_ISDIR(inode->i_mode))
1066 rc = inode_setattr(inode, &op_data->op_attr);
1067 } else if (rc != -EPERM && rc != -EACCES && rc != -ETXTBSY) {
1068 CERROR("md_setattr fails: rc = %d\n", rc);
1073 rc = md_get_lustre_md(sbi->ll_md_exp, request, sbi->ll_dt_exp,
1074 sbi->ll_md_exp, &md);
1076 ptlrpc_req_finished(request);
1080 /* We call inode_setattr to adjust timestamps.
1081 * If there is at least some data in file, we cleared ATTR_SIZE
1082 * above to avoid invoking vmtruncate, otherwise it is important
1083 * to call vmtruncate in inode_setattr to update inode->i_size
1085 rc = inode_setattr(inode, &op_data->op_attr);
1087 /* Extract epoch data if obtained. */
1088 op_data->op_handle = md.body->handle;
1089 op_data->op_ioepoch = md.body->ioepoch;
1091 ll_update_inode(inode, &md);
1092 ptlrpc_req_finished(request);
1097 /* Close IO epoch and send Size-on-MDS attribute update. */
1098 static int ll_setattr_done_writing(struct inode *inode,
1099 struct md_op_data *op_data,
1100 struct md_open_data *mod)
1102 struct ll_inode_info *lli = ll_i2info(inode);
1106 LASSERT(op_data != NULL);
1107 if (!S_ISREG(inode->i_mode))
1110 CDEBUG(D_INODE, "Epoch "LPU64" closed on "DFID" for truncate\n",
1111 op_data->op_ioepoch, PFID(&lli->lli_fid));
1113 op_data->op_flags = MF_EPOCH_CLOSE;
1114 ll_done_writing_attr(inode, op_data);
1115 ll_pack_inode2opdata(inode, op_data, NULL);
1117 rc = md_done_writing(ll_i2sbi(inode)->ll_md_exp, op_data, mod);
1118 if (rc == -EAGAIN) {
1119 /* MDS has instructed us to obtain Size-on-MDS attribute
1120 * from OSTs and send setattr to back to MDS. */
1121 rc = ll_som_update(inode, op_data);
1123 CERROR("inode %lu mdc truncate failed: rc = %d\n",
1129 static int ll_setattr_ost(struct inode *inode, struct iattr *attr)
1131 struct obd_capa *capa;
1134 if (attr->ia_valid & ATTR_SIZE)
1135 capa = ll_osscapa_get(inode, CAPA_OPC_OSS_TRUNC);
1137 capa = ll_mdscapa_get(inode);
1139 rc = cl_setattr_ost(inode, attr, capa);
1141 if (attr->ia_valid & ATTR_SIZE)
1142 ll_truncate_free_capa(capa);
1149 /* If this inode has objects allocated to it (lsm != NULL), then the OST
1150 * object(s) determine the file size and mtime. Otherwise, the MDS will
1151 * keep these values until such a time that objects are allocated for it.
1152 * We do the MDS operations first, as it is checking permissions for us.
1153 * We don't to the MDS RPC if there is nothing that we want to store there,
1154 * otherwise there is no harm in updating mtime/atime on the MDS if we are
1155 * going to do an RPC anyways.
1157 * If we are doing a truncate, we will send the mtime and ctime updates
1158 * to the OST with the punch RPC, otherwise we do an explicit setattr RPC.
1159 * I don't believe it is possible to get e.g. ATTR_MTIME_SET and ATTR_SIZE
1162 int ll_setattr_raw(struct inode *inode, struct iattr *attr)
1164 struct ll_inode_info *lli = ll_i2info(inode);
1165 struct lov_stripe_md *lsm = lli->lli_smd;
1166 struct md_op_data *op_data = NULL;
1167 struct md_open_data *mod = NULL;
1168 int ia_valid = attr->ia_valid;
1169 int rc = 0, rc1 = 0;
1172 CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu valid %x\n", inode->i_ino,
1174 ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_SETATTR, 1);
1176 if (ia_valid & ATTR_SIZE) {
1177 if (attr->ia_size > ll_file_maxbytes(inode)) {
1178 CDEBUG(D_INODE, "file too large %llu > "LPU64"\n",
1179 attr->ia_size, ll_file_maxbytes(inode));
1183 attr->ia_valid |= ATTR_MTIME | ATTR_CTIME;
1186 /* POSIX: check before ATTR_*TIME_SET set (from inode_change_ok) */
1187 if (ia_valid & (ATTR_MTIME_SET | ATTR_ATIME_SET)) {
1188 if (cfs_curproc_fsuid() != inode->i_uid &&
1189 !cfs_capable(CFS_CAP_FOWNER))
1193 /* We mark all of the fields "set" so MDS/OST does not re-set them */
1194 if (attr->ia_valid & ATTR_CTIME) {
1195 attr->ia_ctime = CFS_CURRENT_TIME;
1196 attr->ia_valid |= ATTR_CTIME_SET;
1198 if (!(ia_valid & ATTR_ATIME_SET) && (attr->ia_valid & ATTR_ATIME)) {
1199 attr->ia_atime = CFS_CURRENT_TIME;
1200 attr->ia_valid |= ATTR_ATIME_SET;
1202 if (!(ia_valid & ATTR_MTIME_SET) && (attr->ia_valid & ATTR_MTIME)) {
1203 attr->ia_mtime = CFS_CURRENT_TIME;
1204 attr->ia_valid |= ATTR_MTIME_SET;
1207 if (attr->ia_valid & (ATTR_MTIME | ATTR_CTIME))
1208 CDEBUG(D_INODE, "setting mtime %lu, ctime %lu, now = %lu\n",
1209 LTIME_S(attr->ia_mtime), LTIME_S(attr->ia_ctime),
1210 cfs_time_current_sec());
1212 /* NB: ATTR_SIZE will only be set after this point if the size
1213 * resides on the MDS, ie, this file has no objects. */
1215 attr->ia_valid &= ~ATTR_SIZE;
1217 /* We always do an MDS RPC, even if we're only changing the size;
1218 * only the MDS knows whether truncate() should fail with -ETXTBUSY */
1220 OBD_ALLOC_PTR(op_data);
1221 if (op_data == NULL)
1224 UNLOCK_INODE_MUTEX(inode);
1225 if (ia_valid & ATTR_SIZE)
1226 UP_WRITE_I_ALLOC_SEM(inode);
1227 cfs_down(&lli->lli_trunc_sem);
1228 LOCK_INODE_MUTEX(inode);
1229 if (ia_valid & ATTR_SIZE)
1230 DOWN_WRITE_I_ALLOC_SEM(inode);
1232 memcpy(&op_data->op_attr, attr, sizeof(*attr));
1234 /* Open epoch for truncate. */
1235 if (exp_connect_som(ll_i2mdexp(inode)) &&
1236 (ia_valid & (ATTR_SIZE | ATTR_MTIME | ATTR_MTIME_SET)))
1237 op_data->op_flags = MF_EPOCH_OPEN;
1239 rc = ll_md_setattr(inode, op_data, &mod);
1243 ll_ioepoch_open(lli, op_data->op_ioepoch);
1244 if (!lsm || !S_ISREG(inode->i_mode)) {
1245 CDEBUG(D_INODE, "no lsm: not setting attrs on OST\n");
1249 if (ia_valid & ATTR_SIZE)
1250 attr->ia_valid |= ATTR_SIZE;
1251 if (ia_valid & (ATTR_SIZE | ATTR_MTIME | ATTR_MTIME_SET)) {
1252 /* mtime is set to past sending setattr op to osts under PW
1253 * 0:EOF extent lock (like truncate under PW new_size:EOF), if
1254 * mtime is not set to past setattr op is not sent to osts */
1255 if ((ia_valid & ATTR_SIZE) ||
1256 LTIME_S(attr->ia_mtime) < LTIME_S(attr->ia_ctime))
1257 rc = ll_setattr_ost(inode, attr);
1262 if (op_data->op_ioepoch)
1263 rc1 = ll_setattr_done_writing(inode, op_data, mod);
1264 ll_finish_md_op_data(op_data);
1266 cfs_up(&lli->lli_trunc_sem);
1267 return rc ? rc : rc1;
1270 int ll_setattr(struct dentry *de, struct iattr *attr)
1272 if ((attr->ia_valid & (ATTR_CTIME|ATTR_SIZE|ATTR_MODE)) ==
1273 (ATTR_CTIME|ATTR_SIZE|ATTR_MODE))
1274 attr->ia_valid |= MDS_OPEN_OWNEROVERRIDE;
1276 if ((de->d_inode->i_mode & S_ISUID) &&
1277 !(attr->ia_mode & S_ISUID) &&
1278 !(attr->ia_valid & ATTR_KILL_SUID))
1279 attr->ia_valid |= ATTR_KILL_SUID;
1281 if (((de->d_inode->i_mode & (S_ISGID|S_IXGRP)) == (S_ISGID|S_IXGRP)) &&
1282 !(attr->ia_mode & S_ISGID) &&
1283 !(attr->ia_valid & ATTR_KILL_SGID))
1284 attr->ia_valid |= ATTR_KILL_SGID;
1286 return ll_setattr_raw(de->d_inode, attr);
1289 int ll_statfs_internal(struct super_block *sb, struct obd_statfs *osfs,
1290 __u64 max_age, __u32 flags)
1292 struct ll_sb_info *sbi = ll_s2sbi(sb);
1293 struct obd_statfs obd_osfs;
1297 rc = obd_statfs(class_exp2obd(sbi->ll_md_exp), osfs, max_age, flags);
1299 CERROR("md_statfs fails: rc = %d\n", rc);
1303 osfs->os_type = sb->s_magic;
1305 CDEBUG(D_SUPER, "MDC blocks "LPU64"/"LPU64" objects "LPU64"/"LPU64"\n",
1306 osfs->os_bavail, osfs->os_blocks, osfs->os_ffree,osfs->os_files);
1308 if (sbi->ll_flags & LL_SBI_LAZYSTATFS)
1309 flags |= OBD_STATFS_NODELAY;
1311 rc = obd_statfs_rqset(class_exp2obd(sbi->ll_dt_exp),
1312 &obd_osfs, max_age, flags);
1314 CERROR("obd_statfs fails: rc = %d\n", rc);
1318 CDEBUG(D_SUPER, "OSC blocks "LPU64"/"LPU64" objects "LPU64"/"LPU64"\n",
1319 obd_osfs.os_bavail, obd_osfs.os_blocks, obd_osfs.os_ffree,
1322 osfs->os_bsize = obd_osfs.os_bsize;
1323 osfs->os_blocks = obd_osfs.os_blocks;
1324 osfs->os_bfree = obd_osfs.os_bfree;
1325 osfs->os_bavail = obd_osfs.os_bavail;
1327 /* If we don't have as many objects free on the OST as inodes
1328 * on the MDS, we reduce the total number of inodes to
1329 * compensate, so that the "inodes in use" number is correct.
1331 if (obd_osfs.os_ffree < osfs->os_ffree) {
1332 osfs->os_files = (osfs->os_files - osfs->os_ffree) +
1334 osfs->os_ffree = obd_osfs.os_ffree;
1339 #ifndef HAVE_STATFS_DENTRY_PARAM
1340 int ll_statfs(struct super_block *sb, struct kstatfs *sfs)
1343 int ll_statfs(struct dentry *de, struct kstatfs *sfs)
1345 struct super_block *sb = de->d_sb;
1347 struct obd_statfs osfs;
1350 CDEBUG(D_VFSTRACE, "VFS Op: at "LPU64" jiffies\n", get_jiffies_64());
1351 ll_stats_ops_tally(ll_s2sbi(sb), LPROC_LL_STAFS, 1);
1353 /* For now we will always get up-to-date statfs values, but in the
1354 * future we may allow some amount of caching on the client (e.g.
1355 * from QOS or lprocfs updates). */
1356 rc = ll_statfs_internal(sb, &osfs, cfs_time_current_64() - 1, 0);
1360 statfs_unpack(sfs, &osfs);
1362 /* We need to downshift for all 32-bit kernels, because we can't
1363 * tell if the kernel is being called via sys_statfs64() or not.
1364 * Stop before overflowing f_bsize - in which case it is better
1365 * to just risk EOVERFLOW if caller is using old sys_statfs(). */
1366 if (sizeof(long) < 8) {
1367 while (osfs.os_blocks > ~0UL && sfs->f_bsize < 0x40000000) {
1370 osfs.os_blocks >>= 1;
1371 osfs.os_bfree >>= 1;
1372 osfs.os_bavail >>= 1;
1376 sfs->f_blocks = osfs.os_blocks;
1377 sfs->f_bfree = osfs.os_bfree;
1378 sfs->f_bavail = osfs.os_bavail;
1383 void ll_inode_size_lock(struct inode *inode, int lock_lsm)
1385 struct ll_inode_info *lli;
1386 struct lov_stripe_md *lsm;
1388 lli = ll_i2info(inode);
1389 LASSERT(lli->lli_size_sem_owner != current);
1390 cfs_down(&lli->lli_size_sem);
1391 LASSERT(lli->lli_size_sem_owner == NULL);
1392 lli->lli_size_sem_owner = current;
1394 LASSERTF(lsm != NULL || lock_lsm == 0, "lsm %p, lock_lsm %d\n",
1397 lov_stripe_lock(lsm);
1400 void ll_inode_size_unlock(struct inode *inode, int unlock_lsm)
1402 struct ll_inode_info *lli;
1403 struct lov_stripe_md *lsm;
1405 lli = ll_i2info(inode);
1407 LASSERTF(lsm != NULL || unlock_lsm == 0, "lsm %p, lock_lsm %d\n",
1410 lov_stripe_unlock(lsm);
1411 LASSERT(lli->lli_size_sem_owner == current);
1412 lli->lli_size_sem_owner = NULL;
1413 cfs_up(&lli->lli_size_sem);
1416 void ll_update_inode(struct inode *inode, struct lustre_md *md)
1418 struct ll_inode_info *lli = ll_i2info(inode);
1419 struct mdt_body *body = md->body;
1420 struct lov_stripe_md *lsm = md->lsm;
1421 struct ll_sb_info *sbi = ll_i2sbi(inode);
1423 LASSERT ((lsm != NULL) == ((body->valid & OBD_MD_FLEASIZE) != 0));
1425 cfs_down(&lli->lli_och_sem);
1426 if (lli->lli_smd == NULL) {
1427 if (lsm->lsm_magic != LOV_MAGIC_V1 &&
1428 lsm->lsm_magic != LOV_MAGIC_V3) {
1429 dump_lsm(D_ERROR, lsm);
1432 CDEBUG(D_INODE, "adding lsm %p to inode %lu/%u(%p)\n",
1433 lsm, inode->i_ino, inode->i_generation, inode);
1434 /* cl_inode_init must go before lli_smd or a race is
1435 * possible where client thinks the file has stripes,
1436 * but lov raid0 is not setup yet and parallel e.g.
1437 * glimpse would try to use uninitialized lov */
1438 cl_inode_init(inode, md);
1440 cfs_up(&lli->lli_och_sem);
1441 lli->lli_maxbytes = lsm->lsm_maxbytes;
1442 if (lli->lli_maxbytes > PAGE_CACHE_MAXBYTES)
1443 lli->lli_maxbytes = PAGE_CACHE_MAXBYTES;
1445 cfs_up(&lli->lli_och_sem);
1446 LASSERT(lli->lli_smd->lsm_magic == lsm->lsm_magic &&
1447 lli->lli_smd->lsm_stripe_count ==
1448 lsm->lsm_stripe_count);
1449 if (lov_stripe_md_cmp(lli->lli_smd, lsm)) {
1450 CERROR("lsm mismatch for inode %ld\n",
1452 CERROR("lli_smd:\n");
1453 dump_lsm(D_ERROR, lli->lli_smd);
1455 dump_lsm(D_ERROR, lsm);
1459 if (lli->lli_smd != lsm)
1460 obd_free_memmd(ll_i2dtexp(inode), &lsm);
1463 if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
1464 if (body->valid & OBD_MD_FLRMTPERM)
1465 ll_update_remote_perm(inode, md->remote_perm);
1467 #ifdef CONFIG_FS_POSIX_ACL
1468 else if (body->valid & OBD_MD_FLACL) {
1469 cfs_spin_lock(&lli->lli_lock);
1470 if (lli->lli_posix_acl)
1471 posix_acl_release(lli->lli_posix_acl);
1472 lli->lli_posix_acl = md->posix_acl;
1473 cfs_spin_unlock(&lli->lli_lock);
1476 inode->i_ino = cl_fid_build_ino(&body->fid1);
1477 inode->i_generation = cl_fid_build_gen(&body->fid1);
1479 if (body->valid & OBD_MD_FLATIME) {
1480 if (body->atime > LTIME_S(inode->i_atime))
1481 LTIME_S(inode->i_atime) = body->atime;
1482 lli->lli_lvb.lvb_atime = body->atime;
1484 if (body->valid & OBD_MD_FLMTIME) {
1485 if (body->mtime > LTIME_S(inode->i_mtime)) {
1486 CDEBUG(D_INODE, "setting ino %lu mtime from %lu "
1487 "to "LPU64"\n", inode->i_ino,
1488 LTIME_S(inode->i_mtime), body->mtime);
1489 LTIME_S(inode->i_mtime) = body->mtime;
1491 lli->lli_lvb.lvb_mtime = body->mtime;
1493 if (body->valid & OBD_MD_FLCTIME) {
1494 if (body->ctime > LTIME_S(inode->i_ctime))
1495 LTIME_S(inode->i_ctime) = body->ctime;
1496 lli->lli_lvb.lvb_ctime = body->ctime;
1498 if (body->valid & OBD_MD_FLMODE)
1499 inode->i_mode = (inode->i_mode & S_IFMT)|(body->mode & ~S_IFMT);
1500 if (body->valid & OBD_MD_FLTYPE)
1501 inode->i_mode = (inode->i_mode & ~S_IFMT)|(body->mode & S_IFMT);
1502 LASSERT(inode->i_mode != 0);
1503 if (S_ISREG(inode->i_mode)) {
1504 inode->i_blkbits = min(PTLRPC_MAX_BRW_BITS + 1, LL_MAX_BLKSIZE_BITS);
1506 inode->i_blkbits = inode->i_sb->s_blocksize_bits;
1508 #ifdef HAVE_INODE_BLKSIZE
1509 inode->i_blksize = 1<<inode->i_blkbits;
1511 if (body->valid & OBD_MD_FLUID)
1512 inode->i_uid = body->uid;
1513 if (body->valid & OBD_MD_FLGID)
1514 inode->i_gid = body->gid;
1515 if (body->valid & OBD_MD_FLFLAGS)
1516 inode->i_flags = ll_ext_to_inode_flags(body->flags);
1517 if (body->valid & OBD_MD_FLNLINK)
1518 inode->i_nlink = body->nlink;
1519 if (body->valid & OBD_MD_FLRDEV)
1520 inode->i_rdev = old_decode_dev(body->rdev);
1522 if (body->valid & OBD_MD_FLID) {
1523 /* FID shouldn't be changed! */
1524 if (fid_is_sane(&lli->lli_fid)) {
1525 LASSERTF(lu_fid_eq(&lli->lli_fid, &body->fid1),
1526 "Trying to change FID "DFID
1527 " to the "DFID", inode %lu/%u(%p)\n",
1528 PFID(&lli->lli_fid), PFID(&body->fid1),
1529 inode->i_ino, inode->i_generation, inode);
1531 lli->lli_fid = body->fid1;
1534 LASSERT(fid_seq(&lli->lli_fid) != 0);
1536 if (body->valid & OBD_MD_FLSIZE) {
1537 if (exp_connect_som(ll_i2mdexp(inode)) &&
1538 S_ISREG(inode->i_mode) && lli->lli_smd) {
1539 struct lustre_handle lockh;
1542 /* As it is possible a blocking ast has been processed
1543 * by this time, we need to check there is an UPDATE
1544 * lock on the client and set LLIF_MDS_SIZE_LOCK holding
1546 mode = ll_take_md_lock(inode, MDS_INODELOCK_UPDATE,
1549 if (lli->lli_flags & (LLIF_DONE_WRITING |
1550 LLIF_EPOCH_PENDING |
1552 CERROR("ino %lu flags %lu still has "
1553 "size authority! do not trust "
1554 "the size got from MDS\n",
1555 inode->i_ino, lli->lli_flags);
1557 /* Use old size assignment to avoid
1558 * deadlock bz14138 & bz14326 */
1559 inode->i_size = body->size;
1560 lli->lli_flags |= LLIF_MDS_SIZE_LOCK;
1562 ldlm_lock_decref(&lockh, mode);
1565 /* Use old size assignment to avoid
1566 * deadlock bz14138 & bz14326 */
1567 inode->i_size = body->size;
1570 if (body->valid & OBD_MD_FLBLOCKS)
1571 inode->i_blocks = body->blocks;
1574 if (body->valid & OBD_MD_FLMDSCAPA) {
1575 LASSERT(md->mds_capa);
1576 ll_add_capa(inode, md->mds_capa);
1578 if (body->valid & OBD_MD_FLOSSCAPA) {
1579 LASSERT(md->oss_capa);
1580 ll_add_capa(inode, md->oss_capa);
1584 static struct backing_dev_info ll_backing_dev_info = {
1585 .ra_pages = 0, /* No readahead */
1586 #if (LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,12))
1587 .capabilities = 0, /* Does contribute to dirty memory */
1589 .memory_backed = 0, /* Does contribute to dirty memory */
1593 void ll_read_inode2(struct inode *inode, void *opaque)
1595 struct lustre_md *md = opaque;
1596 struct ll_inode_info *lli = ll_i2info(inode);
1599 CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p)\n",
1600 inode->i_ino, inode->i_generation, inode);
1604 LASSERT(!lli->lli_smd);
1606 /* Core attributes from the MDS first. This is a new inode, and
1607 * the VFS doesn't zero times in the core inode so we have to do
1608 * it ourselves. They will be overwritten by either MDS or OST
1609 * attributes - we just need to make sure they aren't newer. */
1610 LTIME_S(inode->i_mtime) = 0;
1611 LTIME_S(inode->i_atime) = 0;
1612 LTIME_S(inode->i_ctime) = 0;
1614 ll_update_inode(inode, md);
1616 /* OIDEBUG(inode); */
1618 if (S_ISREG(inode->i_mode)) {
1619 struct ll_sb_info *sbi = ll_i2sbi(inode);
1620 inode->i_op = &ll_file_inode_operations;
1621 inode->i_fop = sbi->ll_fop;
1622 inode->i_mapping->a_ops = &ll_aops;
1624 } else if (S_ISDIR(inode->i_mode)) {
1625 inode->i_op = &ll_dir_inode_operations;
1626 inode->i_fop = &ll_dir_operations;
1627 inode->i_mapping->a_ops = &ll_dir_aops;
1629 } else if (S_ISLNK(inode->i_mode)) {
1630 inode->i_op = &ll_fast_symlink_inode_operations;
1633 inode->i_op = &ll_special_inode_operations;
1635 init_special_inode(inode, inode->i_mode,
1636 kdev_t_to_nr(inode->i_rdev));
1638 /* initializing backing dev info. */
1639 inode->i_mapping->backing_dev_info = &ll_backing_dev_info;
1645 void ll_delete_inode(struct inode *inode)
1647 struct ll_sb_info *sbi = ll_i2sbi(inode);
1651 rc = obd_fid_delete(sbi->ll_md_exp, ll_inode2fid(inode));
1653 CERROR("fid_delete() failed, rc %d\n", rc);
1655 truncate_inode_pages(&inode->i_data, 0);
1661 int ll_iocontrol(struct inode *inode, struct file *file,
1662 unsigned int cmd, unsigned long arg)
1664 struct ll_sb_info *sbi = ll_i2sbi(inode);
1665 struct ptlrpc_request *req = NULL;
1670 case FSFILT_IOC_GETFLAGS: {
1671 struct mdt_body *body;
1672 struct md_op_data *op_data;
1674 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL,
1675 0, 0, LUSTRE_OPC_ANY,
1677 if (op_data == NULL)
1680 op_data->op_valid = OBD_MD_FLFLAGS;
1681 rc = md_getattr(sbi->ll_md_exp, op_data, &req);
1682 ll_finish_md_op_data(op_data);
1684 CERROR("failure %d inode %lu\n", rc, inode->i_ino);
1688 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
1690 flags = body->flags;
1692 ptlrpc_req_finished(req);
1694 RETURN(put_user(flags, (int *)arg));
1696 case FSFILT_IOC_SETFLAGS: {
1697 struct lov_stripe_md *lsm = ll_i2info(inode)->lli_smd;
1698 struct obd_info oinfo = { { { 0 } } };
1699 struct md_op_data *op_data;
1701 if (get_user(flags, (int *)arg))
1705 OBDO_ALLOC(oinfo.oi_oa);
1709 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
1710 LUSTRE_OPC_ANY, NULL);
1711 if (IS_ERR(op_data))
1712 RETURN(PTR_ERR(op_data));
1714 ((struct ll_iattr *)&op_data->op_attr)->ia_attr_flags = flags;
1715 op_data->op_attr.ia_valid |= ATTR_ATTR_FLAG;
1716 rc = md_setattr(sbi->ll_md_exp, op_data,
1717 NULL, 0, NULL, 0, &req, NULL);
1718 ll_finish_md_op_data(op_data);
1719 ptlrpc_req_finished(req);
1721 OBDO_FREE(oinfo.oi_oa);
1726 OBDO_FREE(oinfo.oi_oa);
1727 GOTO(update_cache, rc);
1730 oinfo.oi_oa->o_id = lsm->lsm_object_id;
1731 oinfo.oi_oa->o_gr = lsm->lsm_object_gr;
1732 oinfo.oi_oa->o_flags = flags;
1733 oinfo.oi_oa->o_valid = OBD_MD_FLID | OBD_MD_FLFLAGS |
1735 oinfo.oi_capa = ll_mdscapa_get(inode);
1737 obdo_from_inode(oinfo.oi_oa, inode,
1738 OBD_MD_FLFID | OBD_MD_FLGENER);
1739 rc = obd_setattr_rqset(sbi->ll_dt_exp, &oinfo, NULL);
1740 capa_put(oinfo.oi_capa);
1741 OBDO_FREE(oinfo.oi_oa);
1743 if (rc != -EPERM && rc != -EACCES)
1744 CERROR("osc_setattr_async fails: rc = %d\n",rc);
1750 inode->i_flags = ll_ext_to_inode_flags(flags);
1760 int ll_flush_ctx(struct inode *inode)
1762 struct ll_sb_info *sbi = ll_i2sbi(inode);
1764 CDEBUG(D_SEC, "flush context for user %d\n", cfs_curproc_uid());
1766 obd_set_info_async(sbi->ll_md_exp,
1767 sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
1769 obd_set_info_async(sbi->ll_dt_exp,
1770 sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
1775 /* umount -f client means force down, don't save state */
1776 #ifdef HAVE_UMOUNTBEGIN_VFSMOUNT
1777 void ll_umount_begin(struct vfsmount *vfsmnt, int flags)
1779 struct super_block *sb = vfsmnt->mnt_sb;
1781 void ll_umount_begin(struct super_block *sb)
1784 struct lustre_sb_info *lsi = s2lsi(sb);
1785 struct ll_sb_info *sbi = ll_s2sbi(sb);
1786 struct obd_device *obd;
1787 struct obd_ioctl_data *ioc_data;
1790 #ifdef HAVE_UMOUNTBEGIN_VFSMOUNT
1791 if (!(flags & MNT_FORCE)) {
1797 /* Tell the MGC we got umount -f */
1798 lsi->lsi_flags |= LSI_UMOUNT_FORCE;
1800 CDEBUG(D_VFSTRACE, "VFS Op: superblock %p count %d active %d\n", sb,
1801 sb->s_count, atomic_read(&sb->s_active));
1803 obd = class_exp2obd(sbi->ll_md_exp);
1805 CERROR("Invalid MDC connection handle "LPX64"\n",
1806 sbi->ll_md_exp->exp_handle.h_cookie);
1812 obd = class_exp2obd(sbi->ll_dt_exp);
1814 CERROR("Invalid LOV connection handle "LPX64"\n",
1815 sbi->ll_dt_exp->exp_handle.h_cookie);
1821 OBD_ALLOC_PTR(ioc_data);
1823 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_md_exp,
1824 sizeof ioc_data, ioc_data, NULL);
1826 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_dt_exp,
1827 sizeof ioc_data, ioc_data, NULL);
1829 OBD_FREE_PTR(ioc_data);
1833 /* Really, we'd like to wait until there are no requests outstanding,
1834 * and then continue. For now, we just invalidate the requests,
1835 * schedule() and sleep one second if needed, and hope.
1838 #ifdef HAVE_UMOUNTBEGIN_VFSMOUNT
1839 if (atomic_read(&vfsmnt->mnt_count) > 2) {
1840 cfs_schedule_timeout_and_set_state(CFS_TASK_INTERRUPTIBLE,
1841 cfs_time_seconds(1));
1842 if (atomic_read(&vfsmnt->mnt_count) > 2)
1843 LCONSOLE_WARN("Mount still busy with %d refs! You "
1844 "may try to umount it a bit later\n",
1845 atomic_read(&vfsmnt->mnt_count));
1852 int ll_remount_fs(struct super_block *sb, int *flags, char *data)
1854 struct ll_sb_info *sbi = ll_s2sbi(sb);
1858 if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY)) {
1859 read_only = *flags & MS_RDONLY;
1860 err = obd_set_info_async(sbi->ll_md_exp,
1861 sizeof(KEY_READ_ONLY),
1862 KEY_READ_ONLY, sizeof(read_only),
1865 CERROR("Failed to change the read-only flag during "
1866 "remount: %d\n", err);
1871 sb->s_flags |= MS_RDONLY;
1873 sb->s_flags &= ~MS_RDONLY;
1878 int ll_prep_inode(struct inode **inode,
1879 struct ptlrpc_request *req,
1880 struct super_block *sb)
1882 struct ll_sb_info *sbi = NULL;
1883 struct lustre_md md;
1887 LASSERT(*inode || sb);
1888 sbi = sb ? ll_s2sbi(sb) : ll_i2sbi(*inode);
1889 memset(&md, 0, sizeof(struct lustre_md));
1891 rc = md_get_lustre_md(sbi->ll_md_exp, req, sbi->ll_dt_exp,
1892 sbi->ll_md_exp, &md);
1897 ll_update_inode(*inode, &md);
1899 LASSERT(sb != NULL);
1902 * At this point server returns to client's same fid as client
1903 * generated for creating. So using ->fid1 is okay here.
1905 LASSERT(fid_is_sane(&md.body->fid1));
1907 *inode = ll_iget(sb, cl_fid_build_ino(&md.body->fid1), &md);
1908 if (*inode == NULL || IS_ERR(*inode)) {
1910 obd_free_memmd(sbi->ll_dt_exp, &md.lsm);
1911 #ifdef CONFIG_FS_POSIX_ACL
1913 posix_acl_release(md.posix_acl);
1914 md.posix_acl = NULL;
1917 rc = IS_ERR(*inode) ? PTR_ERR(*inode) : -ENOMEM;
1919 CERROR("new_inode -fatal: rc %d\n", rc);
1925 md_free_lustre_md(sbi->ll_md_exp, &md);
1929 int ll_obd_statfs(struct inode *inode, void *arg)
1931 struct ll_sb_info *sbi = NULL;
1932 struct obd_export *exp;
1934 struct obd_ioctl_data *data = NULL;
1938 if (!inode || !(sbi = ll_i2sbi(inode)))
1939 GOTO(out_statfs, rc = -EINVAL);
1941 rc = obd_ioctl_getdata(&buf, &len, arg);
1943 GOTO(out_statfs, rc);
1946 if (!data->ioc_inlbuf1 || !data->ioc_inlbuf2 ||
1947 !data->ioc_pbuf1 || !data->ioc_pbuf2)
1948 GOTO(out_statfs, rc = -EINVAL);
1950 if (data->ioc_inllen1 != sizeof(__u32) ||
1951 data->ioc_inllen2 != sizeof(__u32) ||
1952 data->ioc_plen1 != sizeof(struct obd_statfs) ||
1953 data->ioc_plen2 != sizeof(struct obd_uuid))
1954 GOTO(out_statfs, rc = -EINVAL);
1956 memcpy(&type, data->ioc_inlbuf1, sizeof(__u32));
1957 if (type == LL_STATFS_MDC)
1958 exp = sbi->ll_md_exp;
1959 else if (type == LL_STATFS_LOV)
1960 exp = sbi->ll_dt_exp;
1962 GOTO(out_statfs, rc = -ENODEV);
1964 rc = obd_iocontrol(IOC_OBD_STATFS, exp, len, buf, NULL);
1966 GOTO(out_statfs, rc);
1969 obd_ioctl_freedata(buf, len);
1973 int ll_process_config(struct lustre_cfg *lcfg)
1977 struct lprocfs_static_vars lvars;
1981 lprocfs_llite_init_vars(&lvars);
1983 /* The instance name contains the sb: lustre-client-aacfe000 */
1984 ptr = strrchr(lustre_cfg_string(lcfg, 0), '-');
1985 if (!ptr || !*(++ptr))
1987 if (sscanf(ptr, "%lx", &x) != 1)
1990 /* This better be a real Lustre superblock! */
1991 LASSERT(s2lsi((struct super_block *)sb)->lsi_lmd->lmd_magic == LMD_MAGIC);
1993 /* Note we have not called client_common_fill_super yet, so
1994 proc fns must be able to handle that! */
1995 rc = class_process_proc_param(PARAM_LLITE, lvars.obd_vars,
2002 /* this function prepares md_op_data hint for passing ot down to MD stack. */
2003 struct md_op_data * ll_prep_md_op_data(struct md_op_data *op_data,
2004 struct inode *i1, struct inode *i2,
2005 const char *name, int namelen,
2006 int mode, __u32 opc, void *data)
2008 LASSERT(i1 != NULL);
2010 if (namelen > ll_i2sbi(i1)->ll_namelen)
2011 return ERR_PTR(-ENAMETOOLONG);
2013 if (op_data == NULL)
2014 OBD_ALLOC_PTR(op_data);
2016 if (op_data == NULL)
2017 return ERR_PTR(-ENOMEM);
2019 ll_i2gids(op_data->op_suppgids, i1, i2);
2020 op_data->op_fid1 = *ll_inode2fid(i1);
2021 op_data->op_capa1 = ll_mdscapa_get(i1);
2024 op_data->op_fid2 = *ll_inode2fid(i2);
2025 op_data->op_capa2 = ll_mdscapa_get(i2);
2027 fid_zero(&op_data->op_fid2);
2028 op_data->op_capa2 = NULL;
2031 op_data->op_name = name;
2032 op_data->op_namelen = namelen;
2033 op_data->op_mode = mode;
2034 op_data->op_mod_time = cfs_time_current_sec();
2035 op_data->op_fsuid = cfs_curproc_fsuid();
2036 op_data->op_fsgid = cfs_curproc_fsgid();
2037 op_data->op_cap = cfs_curproc_cap_pack();
2038 op_data->op_bias = MDS_CHECK_SPLIT;
2039 op_data->op_opc = opc;
2040 op_data->op_mds = 0;
2041 op_data->op_data = data;
2046 void ll_finish_md_op_data(struct md_op_data *op_data)
2048 capa_put(op_data->op_capa1);
2049 capa_put(op_data->op_capa2);
2050 OBD_FREE_PTR(op_data);
2053 int ll_show_options(struct seq_file *seq, struct vfsmount *vfs)
2055 struct ll_sb_info *sbi;
2057 LASSERT((seq != NULL) && (vfs != NULL));
2058 sbi = ll_s2sbi(vfs->mnt_sb);
2060 if (sbi->ll_flags & LL_SBI_NOLCK)
2061 seq_puts(seq, ",nolock");
2063 if (sbi->ll_flags & LL_SBI_FLOCK)
2064 seq_puts(seq, ",flock");
2066 if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
2067 seq_puts(seq, ",localflock");
2069 if (sbi->ll_flags & LL_SBI_USER_XATTR)
2070 seq_puts(seq, ",user_xattr");
2072 if (sbi->ll_flags & LL_SBI_ACL)
2073 seq_puts(seq, ",acl");
2075 if (sbi->ll_flags & LL_SBI_LAZYSTATFS)
2076 seq_puts(seq, ",lazystatfs");