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>
48 #include <lustre_lite.h>
49 #include <lustre_ha.h>
50 #include <lustre_dlm.h>
51 #include <lprocfs_status.h>
52 #include <lustre_disk.h>
53 #include <lustre_param.h>
54 #include <lustre_cache.h>
55 #include <obd_support.h>
56 #include "llite_internal.h"
58 cfs_mem_cache_t *ll_file_data_slab;
60 LIST_HEAD(ll_super_blocks);
61 struct rw_semaphore ll_sb_sem;
63 extern struct address_space_operations ll_aops;
64 extern struct address_space_operations ll_dir_aops;
67 #define log2(n) ffz(~(n))
70 static void ll_pglist_fini(struct ll_sb_info *sbi)
75 if (sbi->ll_pglist == NULL)
78 for_each_possible_cpu(i) {
79 page = sbi->ll_pglist[i]->llpd_page;
81 sbi->ll_pglist[i] = NULL;
86 OBD_FREE(sbi->ll_pglist, sizeof(void *)*num_possible_cpus());
87 sbi->ll_pglist = NULL;
90 static int ll_pglist_init(struct ll_sb_info *sbi)
92 struct ll_pglist_data *pd;
97 OBD_ALLOC(sbi->ll_pglist, sizeof(void *) * num_possible_cpus());
98 if (sbi->ll_pglist == NULL)
101 budget = sbi->ll_async_page_max / num_online_cpus();
102 for_each_possible_cpu(i) {
103 struct page *page = alloc_pages_node(cpu_to_node(i),
110 if (color + L1_CACHE_ALIGN(sizeof(*pd)) > PAGE_SIZE)
113 pd = (struct ll_pglist_data *)(page_address(page) + color);
114 memset(pd, 0, sizeof(*pd));
115 spin_lock_init(&pd->llpd_lock);
116 INIT_LIST_HEAD(&pd->llpd_list);
118 pd->llpd_budget = budget;
120 pd->llpd_page = page;
121 atomic_set(&pd->llpd_sample_count, 0);
122 sbi->ll_pglist[i] = pd;
123 color += L1_CACHE_ALIGN(sizeof(*pd));
129 static struct ll_sb_info *ll_init_sbi(void)
131 struct ll_sb_info *sbi = NULL;
138 OBD_ALLOC(sbi, sizeof(*sbi));
142 OBD_ALLOC(sbi->ll_async_page_sample, sizeof(long)*num_possible_cpus());
143 if (sbi->ll_async_page_sample == NULL)
146 spin_lock_init(&sbi->ll_lock);
147 init_mutex(&sbi->ll_lco.lco_lock);
148 spin_lock_init(&sbi->ll_pp_extent_lock);
149 spin_lock_init(&sbi->ll_process_lock);
150 sbi->ll_rw_stats_on = 0;
153 pages = si.totalram - si.totalhigh;
154 if (pages >> (20 - CFS_PAGE_SHIFT) < 512) {
155 #ifdef HAVE_BGL_SUPPORT
156 sbi->ll_async_page_max = pages / 4;
158 sbi->ll_async_page_max = pages / 2;
161 sbi->ll_async_page_max = (pages / 4) * 3;
164 lcounter_init(&sbi->ll_async_page_count);
165 spin_lock_init(&sbi->ll_async_page_reblnc_lock);
166 sbi->ll_async_page_sample_max = 64 * num_online_cpus();
167 sbi->ll_async_page_reblnc_count = 0;
168 sbi->ll_async_page_clock_hand = 0;
169 if (ll_pglist_init(sbi))
172 sbi->ll_ra_info.ra_max_pages_per_file = min(pages / 32,
173 SBI_DEFAULT_READAHEAD_MAX);
174 sbi->ll_ra_info.ra_max_pages = sbi->ll_ra_info.ra_max_pages_per_file;
175 sbi->ll_ra_info.ra_max_read_ahead_whole_pages =
176 SBI_DEFAULT_READAHEAD_WHOLE_MAX;
177 sbi->ll_contention_time = SBI_DEFAULT_CONTENTION_SECONDS;
178 sbi->ll_lockless_truncate_enable = SBI_DEFAULT_LOCKLESS_TRUNCATE_ENABLE;
179 sbi->ll_direct_io_default = SBI_DEFAULT_DIRECT_IO_DEFAULT;
180 INIT_LIST_HEAD(&sbi->ll_conn_chain);
181 INIT_LIST_HEAD(&sbi->ll_orphan_dentry_list);
183 ll_generate_random_uuid(uuid);
184 class_uuid_unparse(uuid, &sbi->ll_sb_uuid);
185 CDEBUG(D_CONFIG, "generated uuid: %s\n", sbi->ll_sb_uuid.uuid);
187 down_write(&ll_sb_sem);
188 list_add_tail(&sbi->ll_list, &ll_super_blocks);
189 up_write(&ll_sb_sem);
191 #ifdef ENABLE_CHECKSUM
192 sbi->ll_flags |= LL_SBI_DATA_CHECKSUM;
194 #ifdef ENABLE_LLITE_CHECKSUM
195 sbi->ll_flags |= LL_SBI_LLITE_CHECKSUM;
198 #ifdef HAVE_LRU_RESIZE_SUPPORT
199 sbi->ll_flags |= LL_SBI_LRU_RESIZE;
202 #ifdef HAVE_EXPORT___IGET
203 INIT_LIST_HEAD(&sbi->ll_deathrow);
204 spin_lock_init(&sbi->ll_deathrow_lock);
206 for (i = 0; i <= LL_PROCESS_HIST_MAX; i++) {
207 spin_lock_init(&sbi->ll_rw_extents_info.pp_extents[i].pp_r_hist.oh_lock);
208 spin_lock_init(&sbi->ll_rw_extents_info.pp_extents[i].pp_w_hist.oh_lock);
211 /* metadata statahead is enabled by default */
212 sbi->ll_sa_max = LL_SA_RPC_DEF;
217 if (sbi->ll_async_page_sample)
218 OBD_FREE(sbi->ll_async_page_sample,
219 sizeof(long) * num_possible_cpus());
221 OBD_FREE(sbi, sizeof(*sbi));
225 void ll_free_sbi(struct super_block *sb)
227 struct ll_sb_info *sbi = ll_s2sbi(sb);
231 down_write(&ll_sb_sem);
232 list_del(&sbi->ll_list);
233 up_write(&ll_sb_sem);
234 /* dont allow find cache via sb list first */
236 lcounter_destroy(&sbi->ll_async_page_count);
237 OBD_FREE(sbi->ll_async_page_sample,
238 sizeof(long) * num_possible_cpus());
239 OBD_FREE(sbi, sizeof(*sbi));
244 static struct dentry_operations ll_d_root_ops = {
245 .d_compare = ll_dcompare,
246 .d_revalidate = ll_revalidate_nd,
249 static int client_common_fill_super(struct super_block *sb,
250 char *mdc, char *osc)
252 struct inode *root = 0;
253 struct ll_sb_info *sbi = ll_s2sbi(sb);
254 struct obd_device *obd;
255 struct ll_fid rootfid;
256 struct obd_statfs osfs;
257 struct ptlrpc_request *request = NULL;
258 struct lustre_handle osc_conn = {0, };
259 struct lustre_handle mdc_conn = {0, };
261 struct obd_connect_data *data = NULL;
265 obd = class_name2obd(mdc);
267 CERROR("MDC %s: not setup or attached\n", mdc);
271 OBD_ALLOC(data, sizeof(*data));
275 if (proc_lustre_fs_root) {
276 err = lprocfs_register_mountpoint(proc_lustre_fs_root, sb,
279 CERROR("could not register mount in /proc/fs/lustre\n");
282 /* indicate the features supported by this client */
283 data->ocd_connect_flags = OBD_CONNECT_VERSION | OBD_CONNECT_IBITS |
284 OBD_CONNECT_JOIN | OBD_CONNECT_ATTRFID |
285 OBD_CONNECT_NODEVOH | OBD_CONNECT_CANCELSET |
286 OBD_CONNECT_AT | OBD_CONNECT_FID |
287 OBD_CONNECT_VBR | OBD_CONNECT_LOV_V3;
288 #ifdef HAVE_LRU_RESIZE_SUPPORT
289 if (sbi->ll_flags & LL_SBI_LRU_RESIZE)
290 data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
292 #ifdef CONFIG_FS_POSIX_ACL
293 data->ocd_connect_flags |= OBD_CONNECT_ACL;
295 data->ocd_ibits_known = MDS_INODELOCK_FULL;
296 data->ocd_version = LUSTRE_VERSION_CODE;
298 if (sb->s_flags & MS_RDONLY)
299 data->ocd_connect_flags |= OBD_CONNECT_RDONLY;
300 if (sbi->ll_flags & LL_SBI_USER_XATTR)
301 data->ocd_connect_flags |= OBD_CONNECT_XATTR;
303 #ifdef HAVE_MS_FLOCK_LOCK
304 /* force vfs to use lustre handler for flock() calls - bug 10743 */
305 sb->s_flags |= MS_FLOCK_LOCK;
308 if (sbi->ll_flags & LL_SBI_FLOCK)
309 sbi->ll_fop = &ll_file_operations_flock;
310 else if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
311 sbi->ll_fop = &ll_file_operations;
313 sbi->ll_fop = &ll_file_operations_noflock;
316 err = obd_connect(&mdc_conn, obd, &sbi->ll_sb_uuid, data, &sbi->ll_mdc_exp);
318 LCONSOLE_ERROR_MSG(0x14f, "An MDT (mdc %s) is performing "
319 "recovery, of which this client is not a "
320 "part. Please wait for recovery to complete,"
321 " abort, or time out.\n", mdc);
324 CERROR("cannot connect to %s: rc = %d\n", mdc, err);
328 err = obd_fid_init(sbi->ll_mdc_exp);
332 err = obd_statfs(obd, &osfs,
333 cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS), 0);
337 /* MDC connect is surely finished by now because we actually sent
338 * a statfs RPC, otherwise obd_connect() is asynchronous. */
339 *data = class_exp2cliimp(sbi->ll_mdc_exp)->imp_connect_data;
341 LASSERT(osfs.os_bsize);
342 sb->s_blocksize = osfs.os_bsize;
343 sb->s_blocksize_bits = log2(osfs.os_bsize);
344 sb->s_magic = LL_SUPER_MAGIC;
346 /* for bug 11559. in $LINUX/fs/read_write.c, function do_sendfile():
347 * retval = in_file->f_op->sendfile(...);
349 * retval = -EOVERFLOW;
351 * it will check if *ppos is greater than max. However, max equals to
352 * s_maxbytes, which is a negative integer in a x86_64 box since loff_t
353 * has been defined as a signed long long ineger in linux kernel. */
354 #if BITS_PER_LONG == 64
355 sb->s_maxbytes = PAGE_CACHE_MAXBYTES >> 1;
357 sb->s_maxbytes = PAGE_CACHE_MAXBYTES;
359 sbi->ll_namelen = osfs.os_namelen;
360 sbi->ll_max_rw_chunk = LL_DEFAULT_MAX_RW_CHUNK;
362 if ((sbi->ll_flags & LL_SBI_USER_XATTR) &&
363 !(data->ocd_connect_flags & OBD_CONNECT_XATTR)) {
364 LCONSOLE_INFO("Disabling user_xattr feature because "
365 "it is not supported on the server\n");
366 sbi->ll_flags &= ~LL_SBI_USER_XATTR;
369 if (data->ocd_connect_flags & OBD_CONNECT_ACL) {
371 sb->s_flags |= MS_POSIXACL;
373 sbi->ll_flags |= LL_SBI_ACL;
375 sbi->ll_flags &= ~LL_SBI_ACL;
377 if (data->ocd_connect_flags & OBD_CONNECT_JOIN)
378 sbi->ll_flags |= LL_SBI_JOIN;
380 obd = class_name2obd(osc);
382 CERROR("OSC %s: not setup or attached\n", osc);
383 GOTO(out_mdc, err = -ENODEV);
386 data->ocd_connect_flags = OBD_CONNECT_VERSION | OBD_CONNECT_GRANT |
387 OBD_CONNECT_REQPORTAL | OBD_CONNECT_BRW_SIZE |
388 OBD_CONNECT_SRVLOCK | OBD_CONNECT_CANCELSET|
389 OBD_CONNECT_AT | OBD_CONNECT_FID |
390 OBD_CONNECT_VBR | OBD_CONNECT_TRUNCLOCK;
392 if (!OBD_FAIL_CHECK(OBD_FAIL_OSC_CONNECT_CKSUM)) {
393 /* OBD_CONNECT_CKSUM should always be set, even if checksums are
394 * disabled by default, because it can still be enabled on the
395 * fly via /proc. As a consequence, we still need to come to an
396 * agreement on the supported algorithms at connect time */
397 data->ocd_connect_flags |= OBD_CONNECT_CKSUM;
399 if (OBD_FAIL_CHECK(OBD_FAIL_OSC_CKSUM_ADLER_ONLY))
400 data->ocd_cksum_types = OBD_CKSUM_ADLER;
402 /* send the list of supported checksum types */
403 data->ocd_cksum_types = OBD_CKSUM_ALL;
406 #ifdef HAVE_LRU_RESIZE_SUPPORT
407 if (sbi->ll_flags & LL_SBI_LRU_RESIZE)
408 data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
411 CDEBUG(D_RPCTRACE, "ocd_connect_flags: "LPX64" ocd_version: %d "
412 "ocd_grant: %d\n", data->ocd_connect_flags,
413 data->ocd_version, data->ocd_grant);
415 obd->obd_upcall.onu_owner = &sbi->ll_lco;
416 obd->obd_upcall.onu_upcall = ll_ocd_update;
417 data->ocd_brw_size = PTLRPC_MAX_BRW_PAGES << CFS_PAGE_SHIFT;
419 /* ask lov to generate OBD_NOTIFY_CREATE events for already registered
421 obd_notify(obd, NULL, OBD_NOTIFY_CREATE, NULL);
423 obd_register_lock_cancel_cb(obd, ll_extent_lock_cancel_cb);
424 obd_register_page_removal_cb(obd, ll_page_removal_cb, ll_pin_extent_cb);
426 err = obd_connect(&osc_conn, obd, &sbi->ll_sb_uuid, data, &sbi->ll_osc_exp);
428 LCONSOLE_ERROR_MSG(0x150, "An OST (osc %s) is performing "
429 "recovery, of which this client is not a "
430 "part. Please wait for recovery to "
431 "complete, abort, or time out.\n", osc);
434 CERROR("cannot connect to %s: rc = %d\n", osc, err);
438 mutex_down(&sbi->ll_lco.lco_lock);
439 sbi->ll_lco.lco_flags = data->ocd_connect_flags;
440 sbi->ll_lco.lco_mdc_exp = sbi->ll_mdc_exp;
441 sbi->ll_lco.lco_osc_exp = sbi->ll_osc_exp;
442 mutex_up(&sbi->ll_lco.lco_lock);
444 err = mdc_init_ea_size(sbi->ll_mdc_exp, sbi->ll_osc_exp);
446 CERROR("cannot set max EA and cookie sizes: rc = %d\n", err);
450 err = obd_prep_async_page(sbi->ll_osc_exp, NULL, NULL, NULL,
451 0, NULL, NULL, NULL, 0, NULL);
453 LCONSOLE_ERROR_MSG(0x151, "There are no OST's in this "
454 "filesystem. There must be at least one "
455 "active OST for a client to start.\n");
459 if (!ll_async_page_slab) {
460 ll_async_page_slab_size =
461 size_round(sizeof(struct ll_async_page)) + err;
462 ll_async_page_slab = cfs_mem_cache_create("ll_async_page",
463 ll_async_page_slab_size,
465 if (!ll_async_page_slab)
466 GOTO(out_osc, err = -ENOMEM);
469 err = mdc_getstatus(sbi->ll_mdc_exp, &rootfid);
471 CERROR("cannot mds_connect: rc = %d\n", err);
474 CDEBUG(D_SUPER, "rootfid "LPU64":"DFID"\n", rootfid.id,
475 PFID((struct lu_fid*)&rootfid));
476 sbi->ll_rootino = rootfid.id;
478 sb->s_op = &lustre_super_operations;
479 #if THREAD_SIZE >= 8192
480 /* Disable the NFS export because of stack overflow
481 * when THREAD_SIZE < 8192. Please refer to 17630. */
482 sb->s_export_op = &lustre_export_operations;
486 * XXX: move this to after cbd setup? */
487 err = mdc_getattr(sbi->ll_mdc_exp, &rootfid,
488 OBD_MD_FLGETATTR | OBD_MD_FLBLOCKS |
489 (sbi->ll_flags & LL_SBI_ACL ? OBD_MD_FLACL : 0),
492 CERROR("mdc_getattr failed for root: rc = %d\n", err);
496 err = mdc_req2lustre_md(request, REPLY_REC_OFF, sbi->ll_osc_exp, &md);
498 CERROR("failed to understand root inode md: rc = %d\n",err);
499 ptlrpc_req_finished (request);
503 LASSERT(sbi->ll_rootino != 0);
504 root = ll_iget(sb, ll_fid_build_ino(sbi, &rootfid), &md);
506 ptlrpc_req_finished(request);
508 if (root == NULL || is_bad_inode(root)) {
509 mdc_free_lustre_md(sbi->ll_osc_exp, &md);
510 CERROR("lustre_lite: bad iget4 for root\n");
511 GOTO(out_root, err = -EBADF);
514 err = ll_close_thread_start(&sbi->ll_lcq);
516 CERROR("cannot start close thread: rc %d\n", err);
520 checksum = sbi->ll_flags & LL_SBI_DATA_CHECKSUM;
521 err = obd_set_info_async(sbi->ll_osc_exp, sizeof(KEY_CHECKSUM),
522 KEY_CHECKSUM, sizeof(checksum),
525 /* making vm readahead 0 for 2.4.x. In the case of 2.6.x,
526 backing dev info assigned to inode mapping is used for
527 determining maximal readahead. */
528 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,6,0)) && \
529 !defined(KERNEL_HAS_AS_MAX_READAHEAD)
530 /* bug 2805 - set VM readahead to zero */
531 vm_max_readahead = vm_min_readahead = 0;
534 sb->s_root = d_alloc_root(root);
536 OBD_FREE(data, sizeof(*data));
537 sb->s_root->d_op = &ll_d_root_ops;
539 sbi->ll_sdev_orig = sb->s_dev;
540 /* We set sb->s_dev equal on all lustre clients in order to support
541 * NFS export clustering. NFSD requires that the FSID be the same
543 /* s_dev is also used in lt_compare() to compare two fs, but that is
544 * only a node-local comparison. */
545 sb->s_dev = get_uuid2int(sbi2mdc(sbi)->cl_target_uuid.uuid,
546 strlen(sbi2mdc(sbi)->cl_target_uuid.uuid));
554 obd_disconnect(sbi->ll_osc_exp);
555 sbi->ll_osc_exp = NULL;
557 obd = class_name2obd(osc);
558 obd_unregister_lock_cancel_cb(obd, ll_extent_lock_cancel_cb);
559 obd_unregister_page_removal_cb(obd, ll_page_removal_cb);
561 obd_fid_fini(sbi->ll_mdc_exp);
562 obd_disconnect(sbi->ll_mdc_exp);
563 sbi->ll_mdc_exp = NULL;
566 OBD_FREE(data, sizeof(*data));
567 lprocfs_unregister_mountpoint(sbi);
571 int ll_get_max_mdsize(struct ll_sb_info *sbi, int *lmmsize)
575 *lmmsize = obd_size_diskmd(sbi->ll_osc_exp, NULL);
577 rc = obd_get_info(sbi->ll_mdc_exp, sizeof(KEY_MAX_EASIZE),
578 KEY_MAX_EASIZE, &size, lmmsize, NULL);
580 CERROR("Get max mdsize error rc %d \n", rc);
585 void ll_dump_inode(struct inode *inode)
587 struct list_head *tmp;
588 int dentry_count = 0;
590 LASSERT(inode != NULL);
592 list_for_each(tmp, &inode->i_dentry)
595 CERROR("inode %p dump: dev=%s ino=%lu mode=%o count=%u, %d dentries\n",
596 inode, ll_i2mdcexp(inode)->exp_obd->obd_name, inode->i_ino,
597 inode->i_mode, atomic_read(&inode->i_count), dentry_count);
600 void lustre_dump_dentry(struct dentry *dentry, int recur)
602 struct list_head *tmp;
605 LASSERT(dentry != NULL);
607 list_for_each(tmp, &dentry->d_subdirs)
610 CERROR("dentry %p dump: name=%.*s parent=%.*s (%p), inode=%p, count=%u,"
611 " flags=0x%x, fsdata=%p, %d subdirs\n", dentry,
612 dentry->d_name.len, dentry->d_name.name,
613 dentry->d_parent->d_name.len, dentry->d_parent->d_name.name,
614 dentry->d_parent, dentry->d_inode, atomic_read(&dentry->d_count),
615 dentry->d_flags, dentry->d_fsdata, subdirs);
616 if (dentry->d_inode != NULL)
617 ll_dump_inode(dentry->d_inode);
622 list_for_each(tmp, &dentry->d_subdirs) {
623 struct dentry *d = list_entry(tmp, struct dentry, d_child);
624 lustre_dump_dentry(d, recur - 1);
628 #ifdef HAVE_EXPORT___IGET
629 static void prune_dir_dentries(struct inode *inode)
631 struct dentry *dentry, *prev = NULL;
633 /* due to lustre specific logic, a directory
634 * can have few dentries - a bug from VFS POV */
636 spin_lock(&dcache_lock);
637 if (!list_empty(&inode->i_dentry)) {
638 dentry = list_entry(inode->i_dentry.prev,
639 struct dentry, d_alias);
640 /* in order to prevent infinite loops we
641 * break if previous dentry is busy */
642 if (dentry != prev) {
645 spin_unlock(&dcache_lock);
647 /* try to kill all child dentries */
648 shrink_dcache_parent(dentry);
651 /* now try to get rid of current dentry */
652 d_prune_aliases(inode);
656 spin_unlock(&dcache_lock);
659 static void prune_deathrow_one(struct ll_inode_info *lli)
661 struct inode *inode = ll_info2i(lli);
663 /* first, try to drop any dentries - they hold a ref on the inode */
664 if (S_ISDIR(inode->i_mode))
665 prune_dir_dentries(inode);
667 d_prune_aliases(inode);
670 /* if somebody still uses it, leave it */
671 LASSERT(atomic_read(&inode->i_count) > 0);
672 if (atomic_read(&inode->i_count) > 1)
675 CDEBUG(D_INODE, "inode %lu/%u(%d) looks a good candidate for prune\n",
676 inode->i_ino,inode->i_generation, atomic_read(&inode->i_count));
678 /* seems nobody uses it anymore */
686 static void prune_deathrow(struct ll_sb_info *sbi, int try)
688 struct ll_inode_info *lli;
692 if (need_resched() && try)
696 if (!spin_trylock(&sbi->ll_deathrow_lock))
699 spin_lock(&sbi->ll_deathrow_lock);
704 if (!list_empty(&sbi->ll_deathrow)) {
705 lli = list_entry(sbi->ll_deathrow.next,
706 struct ll_inode_info,
708 list_del_init(&lli->lli_dead_list);
709 if (!list_empty(&sbi->ll_deathrow))
712 spin_unlock(&sbi->ll_deathrow_lock);
715 prune_deathrow_one(lli);
717 } while (empty == 0);
719 #else /* !HAVE_EXPORT___IGET */
720 #define prune_deathrow(sbi, try) do {} while (0)
721 #endif /* HAVE_EXPORT___IGET */
723 void client_common_put_super(struct super_block *sb)
725 struct ll_sb_info *sbi = ll_s2sbi(sb);
728 ll_close_thread_shutdown(sbi->ll_lcq);
730 lprocfs_unregister_mountpoint(sbi);
732 /* destroy inodes in deathrow */
733 prune_deathrow(sbi, 0);
735 list_del(&sbi->ll_conn_chain);
737 /* callbacks is cleared after disconnect each target */
738 obd_disconnect(sbi->ll_osc_exp);
739 sbi->ll_osc_exp = NULL;
741 obd_fid_fini(sbi->ll_mdc_exp);
742 obd_disconnect(sbi->ll_mdc_exp);
743 sbi->ll_mdc_exp = NULL;
748 void ll_kill_super(struct super_block *sb)
750 struct ll_sb_info *sbi;
755 if (!(sb->s_flags & MS_ACTIVE))
759 /* we need restore s_dev from changed for clustred NFS before put_super
760 * because new kernels have cached s_dev and change sb->s_dev in
761 * put_super not affected real removing devices */
763 sb->s_dev = sbi->ll_sdev_orig;
767 char *ll_read_opt(const char *opt, char *data)
773 CDEBUG(D_SUPER, "option: %s, data %s\n", opt, data);
774 if (strncmp(opt, data, strlen(opt)))
776 if ((value = strchr(data, '=')) == NULL)
780 OBD_ALLOC(retval, strlen(value) + 1);
782 CERROR("out of memory!\n");
786 memcpy(retval, value, strlen(value)+1);
787 CDEBUG(D_SUPER, "Assigned option: %s, value %s\n", opt, retval);
791 static inline int ll_set_opt(const char *opt, char *data, int fl)
793 if (strncmp(opt, data, strlen(opt)) != 0)
799 /* non-client-specific mount options are parsed in lmd_parse */
800 static int ll_options(char *options, int *flags)
803 char *s1 = options, *s2;
809 CDEBUG(D_CONFIG, "Parsing opts %s\n", options);
812 CDEBUG(D_SUPER, "next opt=%s\n", s1);
813 tmp = ll_set_opt("nolock", s1, LL_SBI_NOLCK);
818 tmp = ll_set_opt("flock", s1, LL_SBI_FLOCK);
823 tmp = ll_set_opt("localflock", s1, LL_SBI_LOCALFLOCK);
828 tmp = ll_set_opt("noflock", s1, LL_SBI_FLOCK|LL_SBI_LOCALFLOCK);
833 tmp = ll_set_opt("user_xattr", s1, LL_SBI_USER_XATTR);
838 tmp = ll_set_opt("nouser_xattr", s1, LL_SBI_USER_XATTR);
843 tmp = ll_set_opt("acl", s1, LL_SBI_ACL);
845 /* Ignore deprecated mount option. The client will
846 * always try to mount with ACL support, whether this
847 * is used depends on whether server supports it. */
848 LCONSOLE_ERROR_MSG(0x152, "Ignoring deprecated "
849 "mount option 'acl'.\n");
852 tmp = ll_set_opt("noacl", s1, LL_SBI_ACL);
854 LCONSOLE_ERROR_MSG(0x152, "Ignoring deprecated "
855 "mount option 'noacl'.\n");
859 tmp = ll_set_opt("checksum", s1, LL_SBI_DATA_CHECKSUM);
864 tmp = ll_set_opt("nochecksum", s1, LL_SBI_DATA_CHECKSUM);
870 tmp = ll_set_opt("lruresize", s1, LL_SBI_LRU_RESIZE);
875 tmp = ll_set_opt("nolruresize", s1, LL_SBI_LRU_RESIZE);
880 tmp = ll_set_opt("lazystatfs", s1, LL_SBI_LAZYSTATFS);
885 tmp = ll_set_opt("nolazystatfs", s1, LL_SBI_LAZYSTATFS);
890 LCONSOLE_ERROR_MSG(0x152, "Unknown option '%s', won't mount.\n",
896 s2 = strchr(s1, ',');
904 void ll_lli_init(struct ll_inode_info *lli)
906 lli->lli_inode_magic = LLI_INODE_MAGIC;
907 sema_init(&lli->lli_size_sem, 1);
908 sema_init(&lli->lli_write_sem, 1);
910 lli->lli_maxbytes = PAGE_CACHE_MAXBYTES;
911 spin_lock_init(&lli->lli_lock);
912 sema_init(&lli->lli_och_sem, 1);
913 lli->lli_mds_read_och = lli->lli_mds_write_och = NULL;
914 lli->lli_mds_exec_och = NULL;
915 lli->lli_open_fd_read_count = lli->lli_open_fd_write_count = 0;
916 lli->lli_open_fd_exec_count = 0;
917 INIT_LIST_HEAD(&lli->lli_dead_list);
918 #ifdef HAVE_CLOSE_THREAD
919 INIT_LIST_HEAD(&lli->lli_pending_write_llaps);
921 init_rwsem(&lli->lli_truncate_rwsem);
925 #define MDCDEV "mdc_dev"
926 static int old_lustre_process_log(struct super_block *sb, char *newprofile,
927 struct config_llog_instance *cfg)
929 struct lustre_sb_info *lsi = s2lsi(sb);
930 struct obd_device *obd;
931 struct lustre_handle mdc_conn = {0, };
932 struct obd_export *exp;
933 char *ptr, *mdt, *profile;
934 char niduuid[10] = "mdtnid0";
936 struct obd_uuid mdc_uuid;
937 struct llog_ctxt *ctxt;
938 struct obd_connect_data ocd = { 0 };
940 int i, rc = 0, recov_bk = 1, failnodes = 0;
943 ll_generate_random_uuid(uuid);
944 class_uuid_unparse(uuid, &mdc_uuid);
945 CDEBUG(D_HA, "generated uuid: %s\n", mdc_uuid.uuid);
947 /* Figure out the old mdt and profile name from new-style profile
948 ("lustre" from "mds/lustre-client") */
950 profile = strchr(mdt, '/');
951 if (profile == NULL) {
952 CDEBUG(D_CONFIG, "Can't find MDT name in %s\n", newprofile);
953 GOTO(out, rc = -EINVAL);
957 ptr = strrchr(profile, '-');
959 CDEBUG(D_CONFIG, "Can't find client name in %s\n", newprofile);
960 GOTO(out, rc = -EINVAL);
964 LCONSOLE_WARN("This looks like an old mount command; I will try to "
965 "contact MDT '%s' for profile '%s'\n", mdt, profile);
967 /* Use nids from mount line: uml1,1@elan:uml2,2@elan:/lustre */
969 ptr = lsi->lsi_lmd->lmd_dev;
970 while (class_parse_nid(ptr, &nid, &ptr) == 0) {
971 rc = do_lcfg(MDCDEV, nid, LCFG_ADD_UUID, niduuid, 0,0,0);
973 /* Stop at the first failover nid */
978 CERROR("No valid MDT nids found.\n");
979 GOTO(out, rc = -EINVAL);
983 rc = do_lcfg(MDCDEV, 0, LCFG_ATTACH, LUSTRE_MDC_NAME,mdc_uuid.uuid,0,0);
985 GOTO(out_del_uuid, rc);
987 rc = do_lcfg(MDCDEV, 0, LCFG_SETUP, mdt, niduuid, 0, 0);
989 LCONSOLE_ERROR_MSG(0x153, "I couldn't establish a connection "
990 "with the MDT. Check that the MDT host NID "
991 "is correct and the networks are up.\n");
992 GOTO(out_detach, rc);
995 obd = class_name2obd(MDCDEV);
997 GOTO(out_cleanup, rc = -EINVAL);
999 /* Add any failover nids */
1000 while (*ptr == ':') {
1001 /* New failover node */
1002 sprintf(niduuid, "mdtnid%d", failnodes);
1004 while (class_parse_nid(ptr, &nid, &ptr) == 0) {
1006 rc = do_lcfg(MDCDEV, nid, LCFG_ADD_UUID, niduuid,0,0,0);
1008 CERROR("Add uuid for %s failed %d\n",
1009 libcfs_nid2str(nid), rc);
1014 rc = do_lcfg(MDCDEV, 0, LCFG_ADD_CONN, niduuid, 0, 0,0);
1016 CERROR("Add conn for %s failed %d\n",
1017 libcfs_nid2str(nid), rc);
1025 /* Try all connections, but only once. */
1026 rc = obd_set_info_async(obd->obd_self_export,
1027 sizeof(KEY_INIT_RECOV_BACKUP), KEY_INIT_RECOV_BACKUP,
1028 sizeof(recov_bk), &recov_bk, NULL);
1030 GOTO(out_cleanup, rc);
1032 /* If we don't have this then an ACL MDS will refuse the connection */
1033 ocd.ocd_connect_flags = OBD_CONNECT_ACL;
1035 rc = obd_connect(&mdc_conn, obd, &mdc_uuid, &ocd, &exp);
1037 CERROR("cannot connect to %s: rc = %d\n", mdt, rc);
1038 GOTO(out_cleanup, rc);
1041 ctxt = llog_get_context(exp->exp_obd, LLOG_CONFIG_REPL_CTXT);
1043 cfg->cfg_flags |= CFG_F_COMPAT146;
1046 rc = class_config_parse_llog(ctxt, profile, cfg);
1049 * For debugging, it's useful to just dump the log
1051 rc = class_config_dump_llog(ctxt, profile, cfg);
1053 llog_ctxt_put(ctxt);
1056 /* Set the caller's profile name to the old-style */
1057 memcpy(newprofile, profile, strlen(profile) + 1);
1061 LCONSOLE_ERROR_MSG(0x154, "%s: The configuration '%s' could not"
1062 " be read from the MDT '%s'. Make sure this"
1063 " client and the MDT are running compatible "
1064 "versions of Lustre.\n",
1065 obd->obd_name, profile, mdt);
1068 LCONSOLE_ERROR_MSG(0x155, "%s: The configuration '%s' could not"
1069 " be read from the MDT '%s'. This may be "
1070 "the result of communication errors between "
1071 "the client and the MDT, or if the MDT is "
1072 "not running.\n", obd->obd_name, profile,
1077 /* We don't so much care about errors in cleaning up the config llog
1078 * connection, as we have already read the config by this point. */
1079 obd_disconnect(exp);
1082 do_lcfg(MDCDEV, 0, LCFG_CLEANUP, 0, 0, 0, 0);
1085 do_lcfg(MDCDEV, 0, LCFG_DETACH, 0, 0, 0, 0);
1088 /* class_add_uuid adds a nid even if the same uuid exists; we might
1089 delete any copy here. So they all better match. */
1090 for (i = 0; i < failnodes; i++) {
1091 sprintf(niduuid, "mdtnid%d", i);
1092 do_lcfg(MDCDEV, 0, LCFG_DEL_UUID, niduuid, 0, 0, 0);
1094 /* class_import_put will get rid of the additional connections */
1098 /* end COMPAT_146 */
1100 int ll_fill_super(struct super_block *sb)
1102 struct lustre_profile *lprof = NULL;
1103 struct lustre_sb_info *lsi = s2lsi(sb);
1104 struct ll_sb_info *sbi;
1105 char *osc = NULL, *mdc = NULL;
1106 char *profilenm = get_profile_name(sb);
1107 struct config_llog_instance cfg = {0, };
1108 char ll_instance[sizeof(sb) * 2 + 1];
1111 char pseudo[32] = { 0 };
1114 CDEBUG(D_VFSTRACE, "VFS Op: sb %p\n", sb);
1118 /* client additional sb info */
1119 lsi->lsi_llsbi = sbi = ll_init_sbi();
1125 err = ll_options(lsi->lsi_lmd->lmd_opts, &sbi->ll_flags);
1127 GOTO(out_free, err);
1129 /* Generate a string unique to this super, in case some joker tries
1130 to mount the same fs at two mount points.
1131 Use the address of the super itself.*/
1132 sprintf(ll_instance, "%p", sb);
1133 cfg.cfg_instance = ll_instance;
1134 cfg.cfg_uuid = lsi->lsi_llsbi->ll_sb_uuid;
1137 /* set up client obds */
1138 if (strchr(profilenm, '/') != NULL) /* COMPAT_146 */
1139 err = -EINVAL; /* skip error messages, use old config code */
1141 err = lustre_process_log(sb, profilenm, &cfg);
1146 oldnamelen = strlen(profilenm) + 1;
1147 /* Temp storage for 1.4.6 profile name */
1148 OBD_ALLOC(oldname, oldnamelen);
1150 memcpy(oldname, profilenm, oldnamelen);
1151 rc = old_lustre_process_log(sb, oldname, &cfg);
1153 /* That worked - update the profile name
1156 OBD_FREE(lsi->lsi_lmd->lmd_profile,
1157 strlen(lsi->lsi_lmd->lmd_profile) + 1);
1158 OBD_ALLOC(lsi->lsi_lmd->lmd_profile,
1159 strlen(oldname) + 1);
1160 if (!lsi->lsi_lmd->lmd_profile) {
1161 OBD_FREE(oldname, oldnamelen);
1162 GOTO(out_free, err = -ENOMEM);
1164 memcpy(lsi->lsi_lmd->lmd_profile, oldname,
1165 strlen(oldname) + 1);
1166 profilenm = get_profile_name(sb);
1167 /* Don't ever try to recover the MGS */
1168 rc = ptlrpc_set_import_active(
1169 lsi->lsi_mgc->u.cli.cl_import, 0);
1171 OBD_FREE(oldname, oldnamelen);
1174 /* end COMPAT_146 */
1176 CERROR("Unable to process log: %d\n", err);
1177 GOTO(out_free, err);
1180 lprof = class_get_profile(profilenm);
1181 if (lprof == NULL) {
1182 LCONSOLE_ERROR_MSG(0x156, "The client profile '%s' could not be"
1183 " read from the MGS. Does that filesystem "
1184 "exist?\n", profilenm);
1185 GOTO(out_free, err = -EINVAL);
1189 * The configuration for 1.8 client and 2.0 client are different.
1190 * 2.0 introduces lmv, but 1.8 directly uses mdc.
1191 * Here, we will hack to use proper name for mdc if needed.
1197 save = lprof->lp_mdc;
1198 fsname_end = strrchr(save, '-');
1200 namelen = fsname_end - save;
1201 if (strcmp(fsname_end, "-clilmv") == 0) {
1202 strncpy(pseudo, save, namelen);
1203 strcat(pseudo, "-MDT0000-mdc");
1204 lprof->lp_mdc = pseudo;
1205 CDEBUG(D_INFO, "1.8.x connecting to 2.0: lmv=%s"
1206 " new mdc=%s\n", save, pseudo);
1211 CDEBUG(D_CONFIG, "Found profile %s: mdc=%s osc=%s\n", profilenm,
1212 lprof->lp_mdc, lprof->lp_osc);
1214 OBD_ALLOC(osc, strlen(lprof->lp_osc) +
1215 strlen(ll_instance) + 2);
1217 GOTO(out_free, err = -ENOMEM);
1218 sprintf(osc, "%s-%s", lprof->lp_osc, ll_instance);
1220 OBD_ALLOC(mdc, strlen(lprof->lp_mdc) +
1221 strlen(ll_instance) + 2);
1223 GOTO(out_free, err = -ENOMEM);
1224 sprintf(mdc, "%s-%s", lprof->lp_mdc, ll_instance);
1226 /* connections, registrations, sb setup */
1227 err = client_common_fill_super(sb, mdc, osc);
1231 lprof->lp_mdc = save;
1233 OBD_FREE(mdc, strlen(mdc) + 1);
1235 OBD_FREE(osc, strlen(osc) + 1);
1239 LCONSOLE_WARN("Client %s has started\n", profilenm);
1242 } /* ll_fill_super */
1245 void ll_put_super(struct super_block *sb)
1247 struct config_llog_instance cfg;
1248 char ll_instance[sizeof(sb) * 2 + 1];
1249 struct obd_device *obd;
1250 struct lustre_sb_info *lsi = s2lsi(sb);
1251 struct ll_sb_info *sbi = ll_s2sbi(sb);
1252 char *profilenm = get_profile_name(sb);
1253 int force = 1, next;
1256 CDEBUG(D_VFSTRACE, "VFS Op: sb %p - %s\n", sb, profilenm);
1258 sprintf(ll_instance, "%p", sb);
1259 cfg.cfg_instance = ll_instance;
1260 lustre_end_log(sb, NULL, &cfg);
1262 if (sbi->ll_mdc_exp) {
1263 obd = class_exp2obd(sbi->ll_mdc_exp);
1265 force = obd->obd_force;
1268 /* We need to set force before the lov_disconnect in
1269 lustre_common_put_super, since l_d cleans up osc's as well. */
1272 while ((obd = class_devices_in_group(&sbi->ll_sb_uuid,
1274 obd->obd_force = force;
1279 /* Only if client_common_fill_super succeeded */
1280 client_common_put_super(sb);
1284 while ((obd = class_devices_in_group(&sbi->ll_sb_uuid, &next)) !=NULL) {
1285 class_manual_cleanup(obd);
1289 class_del_profile(profilenm);
1292 lsi->lsi_llsbi = NULL;
1294 lustre_common_put_super(sb);
1296 LCONSOLE_WARN("client %s umount complete\n", ll_instance);
1301 } /* client_put_super */
1303 int ll_shrink_cache(int nr_to_scan, SHRINKER_MASK_T gfp_mask)
1305 struct ll_sb_info *sbi;
1308 if ((nr_to_scan != 0) && (gfp_mask & __GFP_FS))
1311 /* don't race with umount */
1312 down_read(&ll_sb_sem);
1313 list_for_each_entry(sbi, &ll_super_blocks, ll_list)
1314 count += llap_shrink_cache(sbi, nr_to_scan);
1315 up_read(&ll_sb_sem);
1320 struct inode *ll_inode_from_lock(struct ldlm_lock *lock)
1322 struct inode *inode = NULL;
1323 /* NOTE: we depend on atomic igrab() -bzzz */
1324 lock_res_and_lock(lock);
1325 if (lock->l_ast_data) {
1326 struct ll_inode_info *lli = ll_i2info(lock->l_ast_data);
1327 if (lli->lli_inode_magic == LLI_INODE_MAGIC) {
1328 inode = igrab(lock->l_ast_data);
1330 inode = lock->l_ast_data;
1331 ldlm_lock_debug(NULL, inode->i_state & I_FREEING ?
1333 lock, __FILE__, __func__, __LINE__,
1334 "l_ast_data %p is bogus: magic %08x",
1335 lock->l_ast_data, lli->lli_inode_magic);
1339 unlock_res_and_lock(lock);
1343 static int null_if_equal(struct ldlm_lock *lock, void *data)
1345 if (data == lock->l_ast_data) {
1346 lock->l_ast_data = NULL;
1348 if (lock->l_req_mode != lock->l_granted_mode)
1349 LDLM_ERROR(lock,"clearing inode with ungranted lock");
1352 return LDLM_ITER_CONTINUE;
1355 void ll_clear_inode(struct inode *inode)
1358 struct ll_inode_info *lli = ll_i2info(inode);
1359 struct ll_sb_info *sbi = ll_i2sbi(inode);
1362 CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p)\n", inode->i_ino,
1363 inode->i_generation, inode);
1365 if (S_ISDIR(inode->i_mode)) {
1366 /* these should have been cleared in ll_file_release */
1367 LASSERT(lli->lli_sai == NULL);
1368 LASSERT(lli->lli_opendir_key == NULL);
1369 LASSERT(lli->lli_opendir_pid == 0);
1372 ll_inode2fid(&fid, inode);
1373 clear_bit(LLI_F_HAVE_MDS_SIZE_LOCK, &lli->lli_flags);
1374 mdc_change_cbdata(sbi->ll_mdc_exp, &fid, null_if_equal, inode);
1376 LASSERT(!lli->lli_open_fd_write_count);
1377 LASSERT(!lli->lli_open_fd_read_count);
1378 LASSERT(!lli->lli_open_fd_exec_count);
1380 if (lli->lli_mds_write_och)
1381 ll_mdc_real_close(inode, FMODE_WRITE);
1382 if (lli->lli_mds_exec_och) {
1384 CERROR("No FMODE exec, bug exec och is present for "
1385 "inode %ld\n", inode->i_ino);
1386 ll_mdc_real_close(inode, FMODE_EXEC);
1388 if (lli->lli_mds_read_och)
1389 ll_mdc_real_close(inode, FMODE_READ);
1393 obd_change_cbdata(sbi->ll_osc_exp, lli->lli_smd,
1394 null_if_equal, inode);
1396 obd_free_memmd(sbi->ll_osc_exp, &lli->lli_smd);
1397 lli->lli_smd = NULL;
1400 if (lli->lli_symlink_name) {
1401 OBD_FREE(lli->lli_symlink_name,
1402 strlen(lli->lli_symlink_name) + 1);
1403 lli->lli_symlink_name = NULL;
1406 #ifdef CONFIG_FS_POSIX_ACL
1407 if (lli->lli_posix_acl) {
1408 LASSERT(atomic_read(&lli->lli_posix_acl->a_refcount) == 1);
1409 posix_acl_release(lli->lli_posix_acl);
1410 lli->lli_posix_acl = NULL;
1414 lli->lli_inode_magic = LLI_INODE_DEAD;
1416 #ifdef HAVE_EXPORT___IGET
1417 spin_lock(&sbi->ll_deathrow_lock);
1418 list_del_init(&lli->lli_dead_list);
1419 spin_unlock(&sbi->ll_deathrow_lock);
1424 static int ll_setattr_do_truncate(struct inode *inode, loff_t new_size)
1426 struct ll_sb_info *sbi = ll_i2sbi(inode);
1427 struct ll_inode_info *lli = ll_i2info(inode);
1428 struct lov_stripe_md *lsm = lli->lli_smd;
1430 ldlm_policy_data_t policy = { .l_extent = {new_size,
1432 struct lustre_handle lockh = { 0 };
1433 int local_lock = 1; /* 0 - no local lock;
1434 * 1 - lock taken by lock_extent;
1435 * 2 - by obd_match*/
1439 UNLOCK_INODE_MUTEX(inode);
1440 UP_WRITE_I_ALLOC_SEM(inode);
1442 down_write(&lli->lli_truncate_rwsem);
1443 if (sbi->ll_lockless_truncate_enable &&
1444 (sbi->ll_lco.lco_flags & OBD_CONNECT_TRUNCLOCK)) {
1447 ast_flags = LDLM_FL_BLOCK_GRANTED;
1448 rc = obd_match(sbi->ll_osc_exp, lsm, LDLM_EXTENT,
1449 &policy, LCK_PW, &ast_flags, inode, &lockh,
1455 /* clear the lock handle as it not cleared
1456 * by obd_match if no matched lock found */
1458 if (rc == 0 && n_matches == 0) {
1460 rc = ll_file_punch(inode, new_size, 1);
1464 if (local_lock == 1) {
1465 /* XXX when we fix the AST intents to pass the discard-range
1466 * XXX extent, make ast_flags always LDLM_AST_DISCARD_DATA
1468 ast_flags = (new_size == 0) ? LDLM_AST_DISCARD_DATA : 0;
1469 rc = ll_extent_lock(NULL, inode, lsm, LCK_PW, &policy,
1471 if (unlikely(rc != 0))
1475 LOCK_INODE_MUTEX(inode);
1476 DOWN_WRITE_I_ALLOC_SEM(inode);
1477 if (likely(rc == 0)) {
1478 /* Only ll_inode_size_lock is taken at this level.
1479 * lov_stripe_lock() is grabbed by ll_truncate() only over
1480 * call to obd_adjust_kms(). If vmtruncate returns 0, then
1481 * ll_truncate dropped ll_inode_size_lock() */
1482 ll_inode_size_lock(inode, 0);
1484 set_bit(LLI_F_SRVLOCK, &lli->lli_flags);
1485 rc = vmtruncate(inode, new_size);
1486 clear_bit(LLI_F_SRVLOCK, &lli->lli_flags);
1488 LASSERT(SEM_COUNT(&lli->lli_size_sem) <= 0);
1489 ll_inode_size_unlock(inode, 0);
1495 err = (local_lock == 2) ?
1496 obd_cancel(sbi->ll_osc_exp, lsm, LCK_PW, &lockh, 0, 0):
1497 ll_extent_unlock(NULL, inode, lsm, LCK_PW, &lockh);
1498 if (unlikely(err != 0)){
1499 CERROR("extent unlock failed: err=%d,"
1500 " unlock method =%d\n", err, local_lock);
1505 up_write(&lli->lli_truncate_rwsem);
1509 /* If this inode has objects allocated to it (lsm != NULL), then the OST
1510 * object(s) determine the file size and mtime. Otherwise, the MDS will
1511 * keep these values until such a time that objects are allocated for it.
1512 * We do the MDS operations first, as it is checking permissions for us.
1513 * We don't to the MDS RPC if there is nothing that we want to store there,
1514 * otherwise there is no harm in updating mtime/atime on the MDS if we are
1515 * going to do an RPC anyways.
1517 * If we are doing a truncate, we will send the mtime and ctime updates
1518 * to the OST with the punch RPC, otherwise we do an explicit setattr RPC.
1519 * I don't believe it is possible to get e.g. ATTR_MTIME_SET and ATTR_SIZE
1522 int ll_setattr_raw(struct inode *inode, struct iattr *attr)
1524 struct ll_inode_info *lli = ll_i2info(inode);
1525 struct lov_stripe_md *lsm = lli->lli_smd;
1526 struct ll_sb_info *sbi = ll_i2sbi(inode);
1527 struct ptlrpc_request *request = NULL;
1528 struct mdc_op_data *op_data;
1529 struct lustre_md md;
1530 int ia_valid = attr->ia_valid;
1534 CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu valid %x\n", inode->i_ino,
1536 ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_SETATTR, 1);
1538 if (ia_valid & ATTR_SIZE) {
1539 if (attr->ia_size > ll_file_maxbytes(inode)) {
1540 CDEBUG(D_INODE, "file too large %llu > "LPU64"\n",
1541 attr->ia_size, ll_file_maxbytes(inode));
1545 attr->ia_valid |= ATTR_MTIME | ATTR_CTIME;
1548 /* POSIX: check before ATTR_*TIME_SET set (from inode_change_ok) */
1549 if (ia_valid & (ATTR_MTIME_SET | ATTR_ATIME_SET)) {
1550 if (cfs_curproc_fsuid() != inode->i_uid &&
1551 !cfs_capable(CFS_CAP_FOWNER))
1555 /* We mark all of the fields "set" so MDS/OST does not re-set them */
1556 if (attr->ia_valid & ATTR_CTIME) {
1557 attr->ia_ctime = CURRENT_TIME;
1558 attr->ia_valid |= ATTR_CTIME_SET;
1560 if (!(ia_valid & ATTR_ATIME_SET) && (attr->ia_valid & ATTR_ATIME)) {
1561 attr->ia_atime = CURRENT_TIME;
1562 attr->ia_valid |= ATTR_ATIME_SET;
1564 if (!(ia_valid & ATTR_MTIME_SET) && (attr->ia_valid & ATTR_MTIME)) {
1565 attr->ia_mtime = CURRENT_TIME;
1566 attr->ia_valid |= ATTR_MTIME_SET;
1569 if (attr->ia_valid & (ATTR_MTIME | ATTR_CTIME))
1570 CDEBUG(D_INODE, "setting mtime %lu, ctime %lu, now = %lu\n",
1571 LTIME_S(attr->ia_mtime), LTIME_S(attr->ia_ctime),
1574 /* NB: ATTR_SIZE will only be set after this point if the size
1575 * resides on the MDS, ie, this file has no objects. */
1577 attr->ia_valid &= ~ATTR_SIZE;
1579 OBD_ALLOC_PTR(op_data);
1580 if (NULL == op_data)
1582 /* We always do an MDS RPC, even if we're only changing the size;
1583 * only the MDS knows whether truncate() should fail with -ETXTBUSY */
1584 ll_prepare_mdc_op_data(op_data, inode, NULL, NULL, 0, 0, NULL);
1586 rc = mdc_setattr(sbi->ll_mdc_exp, op_data,
1587 attr, NULL, 0, NULL, 0, &request);
1590 ptlrpc_req_finished(request);
1591 if (rc == -ENOENT) {
1593 /* Unlinked special device node? Or just a race?
1594 * Pretend we done everything. */
1595 if (!S_ISREG(inode->i_mode) &&
1596 !S_ISDIR(inode->i_mode))
1597 rc = inode_setattr(inode, attr);
1598 } else if (rc != -EPERM && rc != -EACCES && rc != -ETXTBSY)
1599 CERROR("mdc_setattr fails: rc = %d\n", rc);
1600 OBD_FREE_PTR(op_data);
1603 OBD_FREE_PTR(op_data);
1605 rc = mdc_req2lustre_md(request, REPLY_REC_OFF, sbi->ll_osc_exp, &md);
1607 ptlrpc_req_finished(request);
1611 /* We call inode_setattr to adjust timestamps.
1612 * If there is at least some data in file, we cleared ATTR_SIZE above to
1613 * avoid invoking vmtruncate, otherwise it is important to call
1614 * vmtruncate in inode_setattr to update inode->i_size (bug 6196) */
1615 rc = inode_setattr(inode, attr);
1617 ll_update_inode(inode, &md);
1618 ptlrpc_req_finished(request);
1620 if (!lsm || !S_ISREG(inode->i_mode)) {
1621 CDEBUG(D_INODE, "no lsm: not setting attrs on OST\n");
1625 /* We really need to get our PW lock before we change inode->i_size.
1626 * If we don't we can race with other i_size updaters on our node, like
1627 * ll_file_read. We can also race with i_size propogation to other
1628 * nodes through dirtying and writeback of final cached pages. This
1629 * last one is especially bad for racing o_append users on other
1631 if (ia_valid & ATTR_SIZE) {
1632 rc = ll_setattr_do_truncate(inode, attr->ia_size);
1633 } else if (ia_valid & (ATTR_MTIME | ATTR_MTIME_SET)) {
1634 struct obd_info *oinfo;
1636 struct lustre_handle lockh = { 0 };
1639 OBD_ALLOC_PTR(oinfo);
1643 CDEBUG(D_INODE, "set mtime on OST inode %lu to %lu\n",
1644 inode->i_ino, LTIME_S(attr->ia_mtime));
1648 oa->o_id = lsm->lsm_object_id;
1649 oa->o_gr = lsm->lsm_object_gr;
1650 oa->o_valid = OBD_MD_FLID | OBD_MD_FLGROUP;
1652 valid = OBD_MD_FLTYPE | OBD_MD_FLFID | OBD_MD_FLGENER;
1654 if (LTIME_S(attr->ia_mtime) < LTIME_S(attr->ia_ctime)){
1655 struct ost_lvb xtimes;
1657 /* setting mtime to past is performed under PW
1658 * EOF extent lock */
1659 oinfo->oi_policy.l_extent.start = 0;
1660 oinfo->oi_policy.l_extent.end = OBD_OBJECT_EOF;
1661 rc = ll_extent_lock(NULL, inode, lsm, LCK_PW,
1665 OBD_FREE_PTR(oinfo);
1669 /* setattr under locks
1671 * 1. restore inode's timestamps which
1672 * are about to be set as long as
1673 * concurrent stat (via
1674 * ll_glimpse_size) might bring
1677 * 2. update lsm so that next stat
1678 * (via ll_glimpse_size) could get
1679 * correct values in lsm */
1680 lov_stripe_lock(lli->lli_smd);
1681 if (ia_valid & ATTR_ATIME) {
1682 LTIME_S(inode->i_atime) =
1684 LTIME_S(attr->ia_atime);
1685 valid |= OBD_MD_FLATIME;
1687 if (ia_valid & ATTR_MTIME) {
1688 LTIME_S(inode->i_mtime) =
1690 LTIME_S(attr->ia_mtime);
1691 valid |= OBD_MD_FLMTIME;
1693 if (ia_valid & ATTR_CTIME) {
1694 LTIME_S(inode->i_ctime) =
1696 LTIME_S(attr->ia_ctime);
1697 valid |= OBD_MD_FLCTIME;
1700 obd_update_lvb(ll_i2obdexp(inode), lli->lli_smd,
1702 lov_stripe_unlock(lli->lli_smd);
1706 * 1. do not use inode's timestamps
1707 * because concurrent stat might fill
1708 * the inode with out-of-date times,
1709 * send values from attr instead
1711 * 2.do no update lsm, as long as stat
1712 * (via ll_glimpse_size) will bring
1713 * attributes from osts anyway */
1714 if (ia_valid & ATTR_ATIME) {
1715 oa->o_atime = LTIME_S(attr->ia_atime);
1716 oa->o_valid |= OBD_MD_FLATIME;
1718 if (ia_valid & ATTR_MTIME) {
1719 oa->o_mtime = LTIME_S(attr->ia_mtime);
1720 oa->o_valid |= OBD_MD_FLMTIME;
1722 if (ia_valid & ATTR_CTIME) {
1723 oa->o_ctime = LTIME_S(attr->ia_ctime);
1724 oa->o_valid |= OBD_MD_FLCTIME;
1728 obdo_from_inode(oa, inode, valid);
1733 rc = obd_setattr_rqset(sbi->ll_osc_exp, oinfo, NULL);
1735 CERROR("obd_setattr_async fails: rc=%d\n", rc);
1737 if (LTIME_S(attr->ia_mtime) < LTIME_S(attr->ia_ctime)){
1740 err = ll_extent_unlock(NULL, inode, lsm,
1742 if (unlikely(err != 0)) {
1743 CERROR("extent unlock failed: "
1753 OBD_FREE_PTR(oinfo);
1758 int ll_setattr(struct dentry *de, struct iattr *attr)
1760 int mode = de->d_inode->i_mode;
1762 if ((attr->ia_valid & (ATTR_CTIME|ATTR_SIZE|ATTR_MODE)) ==
1763 (ATTR_CTIME|ATTR_SIZE|ATTR_MODE))
1764 attr->ia_valid |= MDS_OPEN_OWNEROVERRIDE;
1765 if ((attr->ia_valid & (ATTR_MODE|ATTR_FORCE|ATTR_SIZE)) ==
1766 (ATTR_SIZE|ATTR_MODE)) {
1767 if (((mode & S_ISUID) && (!(attr->ia_mode & S_ISUID))) ||
1768 ((mode & S_ISGID) && (mode & S_IXGRP) &&
1769 (!(attr->ia_mode & S_ISGID))))
1770 attr->ia_valid |= ATTR_FORCE;
1773 if ((mode & S_ISUID) && S_ISREG(mode) &&
1774 !(attr->ia_mode & S_ISUID))
1775 attr->ia_valid |= ATTR_KILL_SUID;
1777 if (((mode & (S_ISGID|S_IXGRP)) == (S_ISGID|S_IXGRP)) &&
1778 S_ISREG(mode) && !(attr->ia_mode & S_ISGID))
1779 attr->ia_valid |= ATTR_KILL_SGID;
1781 return ll_setattr_raw(de->d_inode, attr);
1784 int ll_statfs_internal(struct super_block *sb, struct obd_statfs *osfs,
1785 __u64 max_age, __u32 flags)
1787 struct ll_sb_info *sbi = ll_s2sbi(sb);
1788 struct obd_statfs obd_osfs;
1792 rc = obd_statfs(class_exp2obd(sbi->ll_mdc_exp), osfs, max_age, flags);
1794 CERROR("mdc_statfs fails: rc = %d\n", rc);
1798 osfs->os_type = sb->s_magic;
1800 CDEBUG(D_SUPER, "MDC blocks "LPU64"/"LPU64" objects "LPU64"/"LPU64"\n",
1801 osfs->os_bavail, osfs->os_blocks, osfs->os_ffree,osfs->os_files);
1803 if (sbi->ll_flags & LL_SBI_LAZYSTATFS)
1804 flags |= OBD_STATFS_NODELAY;
1806 rc = obd_statfs_rqset(class_exp2obd(sbi->ll_osc_exp),
1807 &obd_osfs, max_age, flags);
1809 CERROR("obd_statfs fails: rc = %d\n", rc);
1813 CDEBUG(D_SUPER, "OSC blocks "LPU64"/"LPU64" objects "LPU64"/"LPU64"\n",
1814 obd_osfs.os_bavail, obd_osfs.os_blocks, obd_osfs.os_ffree,
1817 osfs->os_bsize = obd_osfs.os_bsize;
1818 osfs->os_blocks = obd_osfs.os_blocks;
1819 osfs->os_bfree = obd_osfs.os_bfree;
1820 osfs->os_bavail = obd_osfs.os_bavail;
1822 /* If we don't have as many objects free on the OST as inodes
1823 * on the MDS, we reduce the total number of inodes to
1824 * compensate, so that the "inodes in use" number is correct.
1826 if (obd_osfs.os_ffree < osfs->os_ffree) {
1827 osfs->os_files = (osfs->os_files - osfs->os_ffree) +
1829 osfs->os_ffree = obd_osfs.os_ffree;
1834 #ifndef HAVE_STATFS_DENTRY_PARAM
1835 int ll_statfs(struct super_block *sb, struct kstatfs *sfs)
1838 int ll_statfs(struct dentry *de, struct kstatfs *sfs)
1840 struct super_block *sb = de->d_sb;
1842 struct obd_statfs osfs;
1845 CDEBUG(D_VFSTRACE, "VFS Op: at "LPU64" jiffies\n", get_jiffies_64());
1846 ll_stats_ops_tally(ll_s2sbi(sb), LPROC_LL_STAFS, 1);
1848 /* Some amount of caching on the client is allowed */
1849 rc = ll_statfs_internal(sb, &osfs,
1850 cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS),
1855 statfs_unpack(sfs, &osfs);
1857 /* We need to downshift for all 32-bit kernels, because we can't
1858 * tell if the kernel is being called via sys_statfs64() or not.
1859 * Stop before overflowing f_bsize - in which case it is better
1860 * to just risk EOVERFLOW if caller is using old sys_statfs(). */
1861 if (sizeof(long) < 8) {
1862 while (osfs.os_blocks > ~0UL && sfs->f_bsize < 0x40000000) {
1865 osfs.os_blocks >>= 1;
1866 osfs.os_bfree >>= 1;
1867 osfs.os_bavail >>= 1;
1871 sfs->f_blocks = osfs.os_blocks;
1872 sfs->f_bfree = osfs.os_bfree;
1873 sfs->f_bavail = osfs.os_bavail;
1878 void ll_inode_size_lock(struct inode *inode, int lock_lsm)
1880 struct ll_inode_info *lli;
1881 struct lov_stripe_md *lsm;
1883 lli = ll_i2info(inode);
1884 LASSERT(lli->lli_size_sem_owner != current);
1885 down(&lli->lli_size_sem);
1886 LASSERT(lli->lli_size_sem_owner == NULL);
1887 lli->lli_size_sem_owner = current;
1889 LASSERTF(lsm != NULL || lock_lsm == 0, "lsm %p, lock_lsm %d\n",
1892 lov_stripe_lock(lsm);
1895 void ll_inode_size_unlock(struct inode *inode, int unlock_lsm)
1897 struct ll_inode_info *lli;
1898 struct lov_stripe_md *lsm;
1900 lli = ll_i2info(inode);
1902 LASSERTF(lsm != NULL || unlock_lsm == 0, "lsm %p, lock_lsm %d\n",
1905 lov_stripe_unlock(lsm);
1906 LASSERT(lli->lli_size_sem_owner == current);
1907 lli->lli_size_sem_owner = NULL;
1908 up(&lli->lli_size_sem);
1911 static void ll_replace_lsm(struct inode *inode, struct lov_stripe_md *lsm)
1913 struct ll_inode_info *lli = ll_i2info(inode);
1915 dump_lsm(D_INODE, lsm);
1916 dump_lsm(D_INODE, lli->lli_smd);
1917 LASSERTF(lsm->lsm_magic == LOV_MAGIC_JOIN,
1918 "lsm must be joined lsm %p\n", lsm);
1919 obd_free_memmd(ll_i2obdexp(inode), &lli->lli_smd);
1920 CDEBUG(D_INODE, "replace lsm %p to lli_smd %p for inode %lu%u(%p)\n",
1921 lsm, lli->lli_smd, inode->i_ino, inode->i_generation, inode);
1923 lli->lli_maxbytes = lsm->lsm_maxbytes;
1924 if (lli->lli_maxbytes > PAGE_CACHE_MAXBYTES)
1925 lli->lli_maxbytes = PAGE_CACHE_MAXBYTES;
1928 void ll_update_inode(struct inode *inode, struct lustre_md *md)
1930 struct ll_inode_info *lli = ll_i2info(inode);
1931 struct mds_body *body = md->body;
1932 struct lov_stripe_md *lsm = md->lsm;
1933 struct ll_sb_info *sbi = ll_i2sbi(inode);
1936 CDEBUG(D_INODE, "body->valid = "LPX64"\n", body->valid);
1938 LASSERT ((lsm != NULL) == ((body->valid & OBD_MD_FLEASIZE) != 0));
1940 if (lli->lli_smd == NULL) {
1941 if (lsm->lsm_magic != LOV_MAGIC_V1 &&
1942 lsm->lsm_magic != LOV_MAGIC_V3 &&
1943 lsm->lsm_magic != LOV_MAGIC_JOIN) {
1944 dump_lsm(D_ERROR, lsm);
1947 CDEBUG(D_INODE, "adding lsm %p to inode %lu/%u(%p)\n",
1948 lsm, inode->i_ino, inode->i_generation, inode);
1949 /* ll_inode_size_lock() requires it is only called
1950 * with lli_smd != NULL or lock_lsm == 0 or we can
1951 * race between lock/unlock. bug 9547 */
1953 lli->lli_maxbytes = lsm->lsm_maxbytes;
1954 if (lli->lli_maxbytes > PAGE_CACHE_MAXBYTES)
1955 lli->lli_maxbytes = PAGE_CACHE_MAXBYTES;
1957 if ((lli->lli_smd->lsm_magic == lsm->lsm_magic ||
1958 (lli->lli_smd->lsm_magic == LOV_MAGIC_V3 &&
1959 lsm->lsm_magic == LOV_MAGIC_V1) ||
1960 (lli->lli_smd->lsm_magic == LOV_MAGIC_V1 &&
1961 lsm->lsm_magic == LOV_MAGIC_V3)) &&
1962 lli->lli_smd->lsm_stripe_count ==
1963 lsm->lsm_stripe_count) {
1964 /* The MDS can suddenly change the magic to
1965 * v1/v3 if it has been upgraded/downgraded to a
1966 * version that does/doesn't support OST pools.
1967 * In this case, we consider that the cached lsm
1968 * is equivalent to the new one and keep it in
1969 * place until the inode is evicted from the
1971 if (lov_stripe_md_cmp(lli->lli_smd, lsm)) {
1972 CERROR("lsm mismatch for inode %ld\n",
1974 CERROR("lli_smd:\n");
1975 dump_lsm(D_ERROR, lli->lli_smd);
1977 dump_lsm(D_ERROR, lsm);
1981 ll_replace_lsm(inode, lsm);
1983 if (lli->lli_smd != lsm)
1984 obd_free_memmd(ll_i2obdexp(inode), &lsm);
1987 #ifdef CONFIG_FS_POSIX_ACL
1988 LASSERT(!md->posix_acl || (body->valid & OBD_MD_FLACL));
1989 if (body->valid & OBD_MD_FLACL) {
1990 spin_lock(&lli->lli_lock);
1991 if (lli->lli_posix_acl)
1992 posix_acl_release(lli->lli_posix_acl);
1993 lli->lli_posix_acl = md->posix_acl;
1994 spin_unlock(&lli->lli_lock);
1998 inode->i_ino = ll_fid_build_ino(sbi, &body->fid1);
1999 inode->i_generation = ll_fid_build_gen(sbi, &body->fid1);
2000 *ll_inode_lu_fid(inode) = *((struct lu_fid*)&md->body->fid1);
2002 if (body->valid & OBD_MD_FLATIME) {
2003 if (body->atime > LTIME_S(inode->i_atime))
2004 LTIME_S(inode->i_atime) = body->atime;
2005 lli->lli_lvb.lvb_atime = body->atime;
2007 if (body->valid & OBD_MD_FLMTIME) {
2008 if (body->mtime > LTIME_S(inode->i_mtime)) {
2009 CDEBUG(D_INODE, "setting ino %lu mtime from %lu "
2010 "to "LPU64"\n", inode->i_ino,
2011 LTIME_S(inode->i_mtime), body->mtime);
2012 LTIME_S(inode->i_mtime) = body->mtime;
2014 lli->lli_lvb.lvb_mtime = body->mtime;
2016 if (body->valid & OBD_MD_FLCTIME) {
2017 if (body->ctime > LTIME_S(inode->i_ctime))
2018 LTIME_S(inode->i_ctime) = body->ctime;
2019 lli->lli_lvb.lvb_ctime = body->ctime;
2021 if (body->valid & OBD_MD_FLMODE)
2022 inode->i_mode = (inode->i_mode & S_IFMT)|(body->mode & ~S_IFMT);
2023 if (body->valid & OBD_MD_FLTYPE)
2024 inode->i_mode = (inode->i_mode & ~S_IFMT)|(body->mode & S_IFMT);
2025 if (S_ISREG(inode->i_mode)) {
2026 inode->i_blkbits = min(PTLRPC_MAX_BRW_BITS+1, LL_MAX_BLKSIZE_BITS);
2028 inode->i_blkbits = inode->i_sb->s_blocksize_bits;
2030 #ifdef HAVE_INODE_BLKSIZE
2031 inode->i_blksize = 1<<inode->i_blkbits;
2033 if (body->valid & OBD_MD_FLUID)
2034 inode->i_uid = body->uid;
2035 if (body->valid & OBD_MD_FLGID)
2036 inode->i_gid = body->gid;
2037 if (body->valid & OBD_MD_FLFLAGS)
2038 inode->i_flags = ll_ext_to_inode_flags(body->flags |
2039 MDS_BFLAG_EXT_FLAGS);
2040 if (body->valid & OBD_MD_FLNLINK)
2041 inode->i_nlink = body->nlink;
2043 if (body->valid & OBD_MD_FLRDEV)
2044 inode->i_rdev = old_decode_dev(body->rdev);
2045 if (body->valid & OBD_MD_FLSIZE) {
2046 #if 0 /* Can't block ll_test_inode->ll_update_inode, b=14326*/
2047 ll_inode_size_lock(inode, 0);
2048 i_size_write(inode, body->size);
2049 ll_inode_size_unlock(inode, 0);
2051 inode->i_size = body->size;
2054 if (body->valid & OBD_MD_FLBLOCKS)
2055 inode->i_blocks = body->blocks;
2057 if (body->valid & OBD_MD_FLSIZE)
2058 set_bit(LLI_F_HAVE_MDS_SIZE_LOCK, &lli->lli_flags);
2062 static struct backing_dev_info ll_backing_dev_info = {
2063 .ra_pages = 0, /* No readahead */
2064 #if (LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,12))
2065 .capabilities = 0, /* Does contribute to dirty memory */
2067 .memory_backed = 0, /* Does contribute to dirty memory */
2071 void ll_read_inode2(struct inode *inode, void *opaque)
2073 struct lustre_md *md = opaque;
2074 struct ll_inode_info *lli = ll_i2info(inode);
2077 CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p)\n", inode->i_ino,
2078 inode->i_generation, inode);
2082 LASSERT(!lli->lli_smd);
2084 /* Core attributes from the MDS first. This is a new inode, and
2085 * the VFS doesn't zero times in the core inode so we have to do
2086 * it ourselves. They will be overwritten by either MDS or OST
2087 * attributes - we just need to make sure they aren't newer. */
2088 LTIME_S(inode->i_mtime) = 0;
2089 LTIME_S(inode->i_atime) = 0;
2090 LTIME_S(inode->i_ctime) = 0;
2092 ll_update_inode(inode, md);
2094 /* OIDEBUG(inode); */
2096 if (S_ISREG(inode->i_mode)) {
2097 struct ll_sb_info *sbi = ll_i2sbi(inode);
2098 inode->i_op = &ll_file_inode_operations;
2099 inode->i_fop = sbi->ll_fop;
2100 inode->i_mapping->a_ops = &ll_aops;
2102 } else if (S_ISDIR(inode->i_mode)) {
2103 inode->i_op = &ll_dir_inode_operations;
2104 inode->i_fop = &ll_dir_operations;
2105 inode->i_mapping->a_ops = &ll_dir_aops;
2107 } else if (S_ISLNK(inode->i_mode)) {
2108 inode->i_op = &ll_fast_symlink_inode_operations;
2111 inode->i_op = &ll_special_inode_operations;
2112 init_special_inode(inode, inode->i_mode,
2113 kdev_t_to_nr(inode->i_rdev));
2114 /* initializing backing dev info. */
2115 inode->i_mapping->backing_dev_info = &ll_backing_dev_info;
2120 int ll_iocontrol(struct inode *inode, struct file *file,
2121 unsigned int cmd, unsigned long arg)
2123 struct ll_sb_info *sbi = ll_i2sbi(inode);
2124 struct ptlrpc_request *req = NULL;
2129 case FSFILT_IOC_GETFLAGS: {
2131 struct mds_body *body;
2133 ll_inode2fid(&fid, inode);
2134 rc = mdc_getattr(sbi->ll_mdc_exp, &fid, OBD_MD_FLFLAGS,0,&req);
2136 CERROR("failure %d inode %lu\n", rc, inode->i_ino);
2140 body = lustre_msg_buf(req->rq_repmsg, REPLY_REC_OFF,
2143 /* We want to return EXT3_*_FL flags to the caller via this
2144 * ioctl. An older MDS may be sending S_* flags, fix it up. */
2145 flags = ll_inode_to_ext_flags(body->flags,
2146 MDS_BFLAG_EXT_FLAGS);
2147 ptlrpc_req_finished (req);
2149 RETURN(put_user(flags, (int *)arg));
2151 case FSFILT_IOC_SETFLAGS: {
2152 struct mdc_op_data op_data = { { 0 } };
2153 struct ll_iattr_struct attr;
2154 struct obd_info oinfo = { { { 0 } } };
2155 struct lov_stripe_md *lsm = ll_i2info(inode)->lli_smd;
2157 if (get_user(flags, (int *)arg))
2161 OBDO_ALLOC(oinfo.oi_oa);
2165 ll_prepare_mdc_op_data(&op_data, inode, NULL, NULL, 0, 0, NULL);
2167 memset(&attr, 0, sizeof(attr));
2168 attr.ia_attr_flags = flags;
2169 ((struct iattr *)&attr)->ia_valid |= ATTR_ATTR_FLAG;
2171 rc = mdc_setattr(sbi->ll_mdc_exp, &op_data,
2172 (struct iattr *)&attr, NULL, 0, NULL, 0, &req);
2173 ptlrpc_req_finished(req);
2175 OBDO_FREE(oinfo.oi_oa);
2180 OBDO_FREE(oinfo.oi_oa);
2181 GOTO(update_cache, rc);
2184 oinfo.oi_oa->o_id = lsm->lsm_object_id;
2185 oinfo.oi_oa->o_gr = lsm->lsm_object_gr;
2186 oinfo.oi_oa->o_flags = flags;
2187 oinfo.oi_oa->o_valid = OBD_MD_FLID | OBD_MD_FLGROUP |
2190 obdo_from_inode(oinfo.oi_oa, inode,
2191 OBD_MD_FLFID | OBD_MD_FLGENER);
2192 rc = obd_setattr_rqset(sbi->ll_osc_exp, &oinfo, NULL);
2193 OBDO_FREE(oinfo.oi_oa);
2195 if (rc != -EPERM && rc != -EACCES)
2196 CERROR("osc_setattr_async fails: rc = %d\n",rc);
2202 inode->i_flags = ll_ext_to_inode_flags(flags |
2203 MDS_BFLAG_EXT_FLAGS);
2213 /* umount -f client means force down, don't save state */
2214 #ifdef HAVE_UMOUNTBEGIN_VFSMOUNT
2215 void ll_umount_begin(struct vfsmount *vfsmnt, int flags)
2217 struct super_block *sb = vfsmnt->mnt_sb;
2219 void ll_umount_begin(struct super_block *sb)
2222 struct lustre_sb_info *lsi = s2lsi(sb);
2223 struct ll_sb_info *sbi = ll_s2sbi(sb);
2224 struct obd_device *obd;
2225 struct obd_ioctl_data ioc_data = { 0 };
2228 #ifdef HAVE_UMOUNTBEGIN_VFSMOUNT
2229 if (!(flags & MNT_FORCE)) {
2235 /* Tell the MGC we got umount -f */
2236 lsi->lsi_flags |= LSI_UMOUNT_FORCE;
2238 CDEBUG(D_VFSTRACE, "VFS Op: superblock %p count %d active %d\n", sb,
2239 sb->s_count, atomic_read(&sb->s_active));
2241 obd = class_exp2obd(sbi->ll_mdc_exp);
2243 CERROR("Invalid MDC connection handle "LPX64"\n",
2244 sbi->ll_mdc_exp->exp_handle.h_cookie);
2249 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_mdc_exp, sizeof ioc_data,
2252 obd = class_exp2obd(sbi->ll_osc_exp);
2254 CERROR("Invalid LOV connection handle "LPX64"\n",
2255 sbi->ll_osc_exp->exp_handle.h_cookie);
2261 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_osc_exp, sizeof ioc_data,
2264 /* Really, we'd like to wait until there are no requests outstanding,
2265 * and then continue. For now, we just invalidate the requests,
2266 * schedule() and sleep one second if needed, and hope.
2269 #ifdef HAVE_UMOUNTBEGIN_VFSMOUNT
2270 if (atomic_read(&vfsmnt->mnt_count) > 2) {
2271 cfs_schedule_timeout(CFS_TASK_INTERRUPTIBLE,
2272 cfs_time_seconds(1));
2273 if (atomic_read(&vfsmnt->mnt_count) > 2)
2274 LCONSOLE_WARN("Mount still busy with %d refs! You "
2275 "may try to umount it a bit later\n",
2276 atomic_read(&vfsmnt->mnt_count));
2283 int ll_remount_fs(struct super_block *sb, int *flags, char *data)
2285 struct ll_sb_info *sbi = ll_s2sbi(sb);
2289 if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY)) {
2290 read_only = *flags & MS_RDONLY;
2291 err = obd_set_info_async(sbi->ll_mdc_exp, sizeof(KEY_READONLY),
2292 KEY_READONLY, sizeof(read_only),
2295 /* MDS might have expected a different ro key value, b=17493 */
2296 if (err == -EINVAL) {
2297 CDEBUG(D_CONFIG, "Retrying remount with 1.6.6 ro key\n");
2298 err = obd_set_info_async(sbi->ll_mdc_exp,
2299 sizeof(KEY_READONLY_166COMPAT),
2300 KEY_READONLY_166COMPAT,
2306 CERROR("Failed to change the read-only flag during "
2307 "remount: %d\n", err);
2312 sb->s_flags |= MS_RDONLY;
2314 sb->s_flags &= ~MS_RDONLY;
2319 int ll_prep_inode(struct obd_export *exp, struct inode **inode,
2320 struct ptlrpc_request *req, int offset,struct super_block *sb)
2322 struct lustre_md md;
2323 struct ll_sb_info *sbi = NULL;
2327 LASSERT(*inode || sb);
2328 sbi = sb ? ll_s2sbi(sb) : ll_i2sbi(*inode);
2329 prune_deathrow(sbi, 1);
2331 rc = mdc_req2lustre_md(req, offset, exp, &md);
2336 ll_update_inode(*inode, &md);
2339 /** hashing VFS inode by FIDs.
2340 * IGIF will be used for for compatibility if needed.
2342 *inode =ll_iget(sb, ll_fid_build_ino(sbi, &md.body->fid1), &md);
2343 if (*inode == NULL || is_bad_inode(*inode)) {
2344 mdc_free_lustre_md(exp, &md);
2346 CERROR("new_inode -fatal: rc %d\n", rc);
2351 rc = obd_checkmd(exp, ll_i2mdcexp(*inode),
2352 ll_i2info(*inode)->lli_smd);
2357 char *llap_origins[] = {
2358 [LLAP_ORIGIN_UNKNOWN] = "--",
2359 [LLAP_ORIGIN_READPAGE] = "rp",
2360 [LLAP_ORIGIN_READAHEAD] = "ra",
2361 [LLAP_ORIGIN_COMMIT_WRITE] = "cw",
2362 [LLAP_ORIGIN_WRITEPAGE] = "wp"
2365 struct ll_async_page *llite_pglist_next_llap(struct list_head *head,
2366 struct list_head *list)
2368 struct ll_async_page *llap;
2369 struct list_head *pos;
2371 list_for_each(pos, list) {
2374 llap = list_entry(pos, struct ll_async_page, llap_pglist_item);
2375 if (llap->llap_page == NULL)
2383 int ll_obd_statfs(struct inode *inode, void *arg)
2385 struct ll_sb_info *sbi = NULL;
2386 struct obd_device *client_obd = NULL, *lov_obd = NULL;
2387 struct lov_obd *lov = NULL;
2388 struct obd_statfs stat_buf = {0};
2390 struct obd_ioctl_data *data = NULL;
2394 if (!inode || !(sbi = ll_i2sbi(inode)))
2395 GOTO(out_statfs, rc = -EINVAL);
2397 rc = obd_ioctl_getdata(&buf, &len, arg);
2399 GOTO(out_statfs, rc);
2402 if (!data->ioc_inlbuf1 || !data->ioc_inlbuf2 ||
2403 !data->ioc_pbuf1 || !data->ioc_pbuf2)
2404 GOTO(out_statfs, rc = -EINVAL);
2406 memcpy(&type, data->ioc_inlbuf1, sizeof(__u32));
2407 memcpy(&index, data->ioc_inlbuf2, sizeof(__u32));
2409 if (type == LL_STATFS_MDC) {
2411 GOTO(out_statfs, rc = -ENODEV);
2412 client_obd = class_exp2obd(sbi->ll_mdc_exp);
2413 } else if (type == LL_STATFS_LOV) {
2414 lov_obd = class_exp2obd(sbi->ll_osc_exp);
2415 lov = &lov_obd->u.lov;
2417 if (index >= lov->desc.ld_tgt_count)
2418 GOTO(out_statfs, rc = -ENODEV);
2420 if (!lov->lov_tgts[index])
2421 /* Try again with the next index */
2422 GOTO(out_statfs, rc = -EAGAIN);
2424 client_obd = class_exp2obd(lov->lov_tgts[index]->ltd_exp);
2425 if (!lov->lov_tgts[index]->ltd_active)
2426 GOTO(out_uuid, rc = -ENODATA);
2430 GOTO(out_statfs, rc = -EINVAL);
2432 rc = obd_statfs(client_obd, &stat_buf,
2433 cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS), 1);
2435 GOTO(out_statfs, rc);
2437 if (copy_to_user(data->ioc_pbuf1, &stat_buf, data->ioc_plen1))
2438 GOTO(out_statfs, rc = -EFAULT);
2442 if (copy_to_user(data->ioc_pbuf2, obd2cli_tgt(client_obd),
2446 if (copy_to_user(data->ioc_pbuf2, "Unknown UUID",
2447 sizeof("Unknown UUID") + 1))
2453 obd_ioctl_freedata(buf, len);
2457 int ll_process_config(struct lustre_cfg *lcfg)
2461 struct lprocfs_static_vars lvars;
2465 lprocfs_llite_init_vars(&lvars);
2467 /* The instance name contains the sb: lustre-client-aacfe000 */
2468 ptr = strrchr(lustre_cfg_string(lcfg, 0), '-');
2469 if (!ptr || !*(++ptr))
2471 if (sscanf(ptr, "%lx", &x) != 1)
2474 /* This better be a real Lustre superblock! */
2475 LASSERT(s2lsi((struct super_block *)sb)->lsi_lmd->lmd_magic == LMD_MAGIC);
2477 /* Note we have not called client_common_fill_super yet, so
2478 proc fns must be able to handle that! */
2479 rc = class_process_proc_param(PARAM_LLITE, lvars.obd_vars,
2484 int ll_show_options(struct seq_file *seq, struct vfsmount *vfs)
2486 struct ll_sb_info *sbi;
2488 LASSERT((seq != NULL) && (vfs != NULL));
2489 sbi = ll_s2sbi(vfs->mnt_sb);
2491 if (sbi->ll_flags & LL_SBI_NOLCK)
2492 seq_puts(seq, ",nolock");
2494 if (sbi->ll_flags & LL_SBI_FLOCK)
2495 seq_puts(seq, ",flock");
2497 if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
2498 seq_puts(seq, ",localflock");
2500 if (sbi->ll_flags & LL_SBI_USER_XATTR)
2501 seq_puts(seq, ",user_xattr");
2503 if (sbi->ll_flags & LL_SBI_ACL)
2504 seq_puts(seq, ",acl");
2506 if (sbi->ll_flags & LL_SBI_LAZYSTATFS)
2507 seq_puts(seq, ",lazystatfs");