4 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
6 * This program is free software; you can redistribute it and/or modify
7 * it under the terms of the GNU General Public License version 2 only,
8 * as published by the Free Software Foundation.
10 * This program is distributed in the hope that it will be useful, but
11 * WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
13 * General Public License version 2 for more details (a copy is included
14 * in the LICENSE file that accompanied this code).
16 * You should have received a copy of the GNU General Public License
17 * version 2 along with this program; If not, see
18 * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
20 * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
21 * CA 95054 USA or visit www.sun.com if you need additional information or
27 * Copyright (c) 2002, 2010, Oracle and/or its affiliates. All rights reserved.
28 * Use is subject to license terms.
30 * Copyright (c) 2011, 2013, Intel Corporation.
33 * This file is part of Lustre, http://www.lustre.org/
34 * Lustre is a trademark of Sun Microsystems, Inc.
38 * Directory code for lustre client.
42 #include <linux/pagemap.h>
44 #include <linux/version.h>
45 #include <asm/uaccess.h>
46 #include <linux/buffer_head.h> // for wait_on_buffer
47 #include <linux/pagevec.h>
49 #define DEBUG_SUBSYSTEM S_LLITE
51 #include <lustre/lustre_idl.h>
52 #include <obd_support.h>
53 #include <obd_class.h>
54 #include <lustre_lib.h>
55 #include <lustre/lustre_idl.h>
56 #include <lustre_lite.h>
57 #include <lustre_dlm.h>
58 #include <lustre_fid.h>
59 #include "llite_internal.h"
62 * (new) readdir implementation overview.
64 * Original lustre readdir implementation cached exact copy of raw directory
65 * pages on the client. These pages were indexed in client page cache by
66 * logical offset in the directory file. This design, while very simple and
67 * intuitive had some inherent problems:
69 * . it implies that byte offset to the directory entry serves as a
70 * telldir(3)/seekdir(3) cookie, but that offset is not stable: in
71 * ext3/htree directory entries may move due to splits, and more
74 * . it is incompatible with the design of split directories for cmd3,
75 * that assumes that names are distributed across nodes based on their
76 * hash, and so readdir should be done in hash order.
78 * New readdir implementation does readdir in hash order, and uses hash of a
79 * file name as a telldir/seekdir cookie. This led to number of complications:
81 * . hash is not unique, so it cannot be used to index cached directory
82 * pages on the client (note, that it requires a whole pageful of hash
83 * collided entries to cause two pages to have identical hashes);
85 * . hash is not unique, so it cannot, strictly speaking, be used as an
86 * entry cookie. ext3/htree has the same problem and lustre implementation
87 * mimics their solution: seekdir(hash) positions directory at the first
88 * entry with the given hash.
94 * Client caches directory pages using hash of the first entry as an index. As
95 * noted above hash is not unique, so this solution doesn't work as is:
96 * special processing is needed for "page hash chains" (i.e., sequences of
97 * pages filled with entries all having the same hash value).
99 * First, such chains have to be detected. To this end, server returns to the
100 * client the hash of the first entry on the page next to one returned. When
101 * client detects that this hash is the same as hash of the first entry on the
102 * returned page, page hash collision has to be handled. Pages in the
103 * hash chain, except first one, are termed "overflow pages".
105 * Solution to index uniqueness problem is to not cache overflow
106 * pages. Instead, when page hash collision is detected, all overflow pages
107 * from emerging chain are immediately requested from the server and placed in
108 * a special data structure (struct ll_dir_chain). This data structure is used
109 * by ll_readdir() to process entries from overflow pages. When readdir
110 * invocation finishes, overflow pages are discarded. If page hash collision
111 * chain weren't completely processed, next call to readdir will again detect
112 * page hash collision, again read overflow pages in, process next portion of
113 * entries and again discard the pages. This is not as wasteful as it looks,
114 * because, given reasonable hash, page hash collisions are extremely rare.
116 * 1. directory positioning
118 * When seekdir(hash) is called, original
129 * identification of and access to overflow pages
133 * Page in MDS_READPAGE RPC is packed in LU_PAGE_SIZE, and each page contains
134 * a header lu_dirpage which describes the start/end hash, and whether this
135 * page is empty (contains no dir entry) or hash collide with next page.
136 * After client receives reply, several pages will be integrated into dir page
137 * in PAGE_CACHE_SIZE (if PAGE_CACHE_SIZE greater than LU_PAGE_SIZE), and the
138 * lu_dirpage for this integrated page will be adjusted. See
139 * lmv_adjust_dirpages().
143 * The following three APIs will be used by llite to iterate directory
144 * entries from MDC dir page caches.
146 * ll_dir_entry_start(next) will lookup(return) entry by op_hash_offset.
147 * To avoid extra memory allocation, the @entry will be pointed to
148 * the dir entries in MDC page directly, so these pages can not be released
149 * until the entry has been accessed in ll_readdir(or statahead).
151 * The iterate process will be
153 * ll_dir_entry_start: locate the page in MDC, and return the first entry.
156 * ll_dir_entry_next: return the next entry in the current page, if it reaches
157 * to the end, release current page.
159 * ll_dir_entry_end: release the last page.
161 struct lu_dirent *ll_dir_entry_start(struct inode *dir,
162 struct md_op_data *op_data,
165 struct lu_dirent *entry;
166 struct md_callback cb_op;
169 LASSERT(*ppage == NULL);
170 cb_op.md_blocking_ast = ll_md_blocking_ast;
171 rc = md_read_entry(ll_i2mdexp(dir), op_data, &cb_op, &entry, ppage);
177 struct lu_dirent *ll_dir_entry_next(struct inode *dir,
178 struct md_op_data *op_data,
179 struct lu_dirent *ent,
182 struct lu_dirent *entry;
183 struct md_callback cb_op;
186 LASSERT(*ppage != NULL);
187 cb_op.md_blocking_ast = ll_md_blocking_ast;
188 op_data->op_hash_offset = le64_to_cpu(ent->lde_hash);
190 page_cache_release(*ppage);
192 rc = md_read_entry(ll_i2mdexp(dir), op_data, &cb_op, &entry, ppage);
198 int ll_dir_read(struct inode *inode, struct md_op_data *op_data,
199 void *cookie, filldir_t filldir)
201 struct ll_sb_info *sbi = ll_i2sbi(inode);
202 struct ll_dir_chain chain;
203 struct lu_dirent *ent;
204 int api32 = ll_need_32bit_api(sbi);
205 int hash64 = sbi->ll_flags & LL_SBI_64BIT_HASH;
208 __u64 hash = MDS_DIR_END_OFF;
209 __u64 last_hash = MDS_DIR_END_OFF;
210 struct page *page = NULL;
213 ll_dir_chain_init(&chain);
214 for (ent = ll_dir_entry_start(inode, op_data, &page);
215 ent != NULL && !IS_ERR(ent) && !done;
216 ent = ll_dir_entry_next(inode, op_data, ent, &page)) {
223 hash = le64_to_cpu(ent->lde_hash);
224 if (hash < op_data->op_hash_offset)
226 * Skip until we find target hash
230 namelen = le16_to_cpu(ent->lde_namelen);
241 fid_le_to_cpu(&fid, &ent->lde_fid);
242 ino = cl_fid_build_ino(&fid, api32);
243 type = ll_dirent_type_get(ent);
244 /* For 'll_nfs_get_name_filldir()', it will try
245 * to access the 'ent' through its 'lde_name',
246 * so the parameter 'name' for 'filldir()' must
247 * be part of the 'ent'. */
248 done = filldir(cookie, ent->lde_name, namelen, lhash,
251 if (op_data->op_hash_offset != MDS_DIR_END_OFF)
252 op_data->op_hash_offset = last_hash;
264 page_cache_release(page);
267 ll_dir_chain_fini(&chain);
271 static int ll_readdir(struct file *filp, void *cookie, filldir_t filldir)
273 struct inode *inode = filp->f_dentry->d_inode;
274 struct ll_file_data *lfd = LUSTRE_FPRIVATE(filp);
275 struct ll_sb_info *sbi = ll_i2sbi(inode);
276 int hash64 = sbi->ll_flags & LL_SBI_64BIT_HASH;
277 int api32 = ll_need_32bit_api(sbi);
278 struct md_op_data *op_data;
281 #ifdef HAVE_TOUCH_ATIME_1ARG
291 CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p) pos/size"
292 "%lu/%llu 32bit_api %d\n", PFID(ll_inode2fid(inode)),
293 inode, (unsigned long)pos, i_size_read(inode), api32);
295 if (pos == MDS_DIR_END_OFF)
301 op_data = ll_prep_md_op_data(NULL, inode, inode, NULL, 0, 0,
302 LUSTRE_OPC_ANY, inode);
304 GOTO(out, rc = PTR_ERR(op_data));
306 op_data->op_hash_offset = pos;
307 op_data->op_max_pages = sbi->ll_md_brw_size >> PAGE_CACHE_SHIFT;
308 rc = ll_dir_read(inode, op_data, cookie, filldir);
310 lfd->lfd_pos = op_data->op_hash_offset;
312 if (pos == MDS_DIR_END_OFF) {
314 filp->f_pos = LL_DIR_END_OFF_32BIT;
316 filp->f_pos = LL_DIR_END_OFF;
319 filp->f_pos = op_data->op_hash_offset >> 32;
321 filp->f_pos = op_data->op_hash_offset;
324 ll_finish_md_op_data(op_data);
325 filp->f_version = inode->i_version;
326 #ifdef HAVE_TOUCH_ATIME_1ARG
327 #ifdef HAVE_F_PATH_MNT
328 path.mnt = filp->f_path.mnt;
330 path.mnt = filp->f_vfsmnt;
332 path.dentry = filp->f_dentry;
335 touch_atime(filp->f_vfsmnt, filp->f_dentry);
340 ll_stats_ops_tally(sbi, LPROC_LL_READDIR, 1);
345 int ll_send_mgc_param(struct obd_export *mgc, char *string)
347 struct mgs_send_param *msp;
354 strncpy(msp->mgs_param, string, MGS_PARAM_MAXLEN);
355 rc = obd_set_info_async(NULL, mgc, sizeof(KEY_SET_INFO), KEY_SET_INFO,
356 sizeof(struct mgs_send_param), msp, NULL);
358 CERROR("Failed to set parameter: %d\n", rc);
364 static int ll_dir_setdirstripe(struct inode *dir, struct lmv_user_md *lump,
365 const char *filename)
367 struct ptlrpc_request *request = NULL;
368 struct md_op_data *op_data;
369 struct ll_sb_info *sbi = ll_i2sbi(dir);
374 if (unlikely(lump->lum_magic != LMV_USER_MAGIC))
377 CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p) name %s"
378 "stripe_offset %d, stripe_count: %u\n",
379 PFID(ll_inode2fid(dir)), dir, filename,
380 (int)lump->lum_stripe_offset, lump->lum_stripe_count);
382 if (lump->lum_magic != cpu_to_le32(LMV_USER_MAGIC))
383 lustre_swab_lmv_user_md(lump);
385 mode = (0755 & (S_IRWXUGO|S_ISVTX) & ~current->fs->umask) | S_IFDIR;
386 op_data = ll_prep_md_op_data(NULL, dir, NULL, filename,
387 strlen(filename), mode, LUSTRE_OPC_MKDIR,
390 GOTO(err_exit, err = PTR_ERR(op_data));
392 op_data->op_cli_flags |= CLI_SET_MEA;
393 err = md_create(sbi->ll_md_exp, op_data, lump, sizeof(*lump), mode,
394 current_fsuid(), current_fsgid(),
395 cfs_curproc_cap_pack(), 0, &request);
396 ll_finish_md_op_data(op_data);
400 ptlrpc_req_finished(request);
404 int ll_dir_setstripe(struct inode *inode, struct lov_user_md *lump,
407 struct ll_sb_info *sbi = ll_i2sbi(inode);
408 struct md_op_data *op_data;
409 struct ptlrpc_request *req = NULL;
411 struct lustre_sb_info *lsi = s2lsi(inode->i_sb);
412 struct obd_device *mgc = lsi->lsi_mgc;
418 * This is coming from userspace, so should be in
419 * local endian. But the MDS would like it in little
420 * endian, so we swab it before we send it.
422 switch (lump->lmm_magic) {
423 case LOV_USER_MAGIC_V1: {
424 if (lump->lmm_magic != cpu_to_le32(LOV_USER_MAGIC_V1))
425 lustre_swab_lov_user_md_v1(lump);
426 lum_size = sizeof(struct lov_user_md_v1);
429 case LOV_USER_MAGIC_V3: {
430 if (lump->lmm_magic != cpu_to_le32(LOV_USER_MAGIC_V3))
431 lustre_swab_lov_user_md_v3(
432 (struct lov_user_md_v3 *)lump);
433 lum_size = sizeof(struct lov_user_md_v3);
436 case LMV_USER_MAGIC: {
437 if (lump->lmm_magic != cpu_to_le32(LMV_USER_MAGIC))
438 lustre_swab_lmv_user_md(
439 (struct lmv_user_md *)lump);
440 lum_size = sizeof(struct lmv_user_md);
444 CDEBUG(D_IOCTL, "bad userland LOV MAGIC:"
445 " %#08x != %#08x nor %#08x\n",
446 lump->lmm_magic, LOV_USER_MAGIC_V1,
452 lum_size = sizeof(struct lov_user_md_v1);
455 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
456 LUSTRE_OPC_ANY, NULL);
458 RETURN(PTR_ERR(op_data));
460 /* swabbing is done in lov_setstripe() on server side */
461 rc = md_setattr(sbi->ll_md_exp, op_data, lump, lum_size,
462 NULL, 0, &req, NULL);
463 ll_finish_md_op_data(op_data);
464 ptlrpc_req_finished(req);
466 if (rc != -EPERM && rc != -EACCES)
467 CERROR("mdc_setattr fails: rc = %d\n", rc);
470 /* In the following we use the fact that LOV_USER_MAGIC_V1 and
471 LOV_USER_MAGIC_V3 have the same initial fields so we do not
472 need the make the distiction between the 2 versions */
473 if (set_default && mgc->u.cli.cl_mgc_mgsexp) {
477 OBD_ALLOC(param, MGS_PARAM_MAXLEN);
479 GOTO(end, rc = -ENOMEM);
482 /* Get fsname and assume devname to be -MDT0000. */
483 ll_get_fsname(inode->i_sb, buf, MTI_NAME_MAXLEN);
484 strcat(buf, "-MDT0000.lov");
487 /* Set root stripesize */
488 sprintf(buf, ".stripesize=%u",
489 lump ? le32_to_cpu(lump->lmm_stripe_size) : 0);
490 rc = ll_send_mgc_param(mgc->u.cli.cl_mgc_mgsexp, param);
494 /* Set root stripecount */
495 sprintf(buf, ".stripecount=%hd",
496 lump ? le16_to_cpu(lump->lmm_stripe_count) : 0);
497 rc = ll_send_mgc_param(mgc->u.cli.cl_mgc_mgsexp, param);
501 /* Set root stripeoffset */
502 sprintf(buf, ".stripeoffset=%hd",
503 lump ? le16_to_cpu(lump->lmm_stripe_offset) :
504 (typeof(lump->lmm_stripe_offset))(-1));
505 rc = ll_send_mgc_param(mgc->u.cli.cl_mgc_mgsexp, param);
509 OBD_FREE(param, MGS_PARAM_MAXLEN);
515 * This function will be used to get default LOV/LMV/Default LMV
516 * @valid will be used to indicate which stripe it will retrieve
517 * OBD_MD_MEA LMV stripe EA
518 * OBD_MD_DEFAULT_MEA Default LMV stripe EA
519 * otherwise Default LOV EA.
520 * Each time, it can only retrieve 1 stripe EA
522 int ll_dir_getstripe(struct inode *inode, void **plmm, int *plmm_size,
523 struct ptlrpc_request **request, obd_valid valid)
525 struct ll_sb_info *sbi = ll_i2sbi(inode);
526 struct mdt_body *body;
527 struct lov_mds_md *lmm = NULL;
528 struct ptlrpc_request *req = NULL;
530 struct md_op_data *op_data;
533 rc = ll_get_max_mdsize(sbi, &lmm_size);
537 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL,
538 0, lmm_size, LUSTRE_OPC_ANY,
541 RETURN(PTR_ERR(op_data));
543 op_data->op_valid = valid | OBD_MD_FLEASIZE | OBD_MD_FLDIREA;
544 rc = md_getattr(sbi->ll_md_exp, op_data, &req);
545 ll_finish_md_op_data(op_data);
547 CDEBUG(D_INFO, "md_getattr failed on inode "
548 DFID": rc %d\n", PFID(ll_inode2fid(inode)), rc);
552 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
553 LASSERT(body != NULL);
555 lmm_size = body->eadatasize;
557 if (!(body->valid & (OBD_MD_FLEASIZE | OBD_MD_FLDIREA)) ||
559 GOTO(out, rc = -ENODATA);
562 lmm = req_capsule_server_sized_get(&req->rq_pill,
563 &RMF_MDT_MD, lmm_size);
564 LASSERT(lmm != NULL);
567 * This is coming from the MDS, so is probably in
568 * little endian. We convert it to host endian before
569 * passing it to userspace.
571 /* We don't swab objects for directories */
572 switch (le32_to_cpu(lmm->lmm_magic)) {
574 if (LOV_MAGIC != cpu_to_le32(LOV_MAGIC))
575 lustre_swab_lov_user_md_v1((struct lov_user_md_v1 *)lmm);
578 if (LOV_MAGIC != cpu_to_le32(LOV_MAGIC))
579 lustre_swab_lov_user_md_v3((struct lov_user_md_v3 *)lmm);
582 if (LOV_MAGIC != cpu_to_le32(LOV_MAGIC))
583 lustre_swab_lmv_mds_md((union lmv_mds_md *)lmm);
586 if (LMV_USER_MAGIC != cpu_to_le32(LMV_USER_MAGIC))
587 lustre_swab_lmv_user_md((struct lmv_user_md *)lmm);
590 CERROR("unknown magic: %lX\n", (unsigned long)lmm->lmm_magic);
595 *plmm_size = lmm_size;
600 static int ll_get_mdt_idx_by_fid(struct ll_sb_info *sbi,
601 const struct lu_fid *fid)
603 struct md_op_data *op_data;
608 OBD_ALLOC_PTR(op_data);
612 op_data->op_flags |= MF_GET_MDT_IDX;
613 op_data->op_fid1 = *fid;
614 rc = md_getattr(sbi->ll_md_exp, op_data, NULL);
615 mdt_index = op_data->op_mds;
616 OBD_FREE_PTR(op_data);
624 * Get MDT index for the inode.
626 int ll_get_mdt_idx(struct inode *inode)
628 return ll_get_mdt_idx_by_fid(ll_i2sbi(inode), ll_inode2fid(inode));
632 * Generic handler to do any pre-copy work.
634 * It send a first hsm_progress (with extent length == 0) to coordinator as a
635 * first information for it that real work has started.
637 * Moreover, for a ARCHIVE request, it will sample the file data version and
638 * store it in \a copy.
640 * \return 0 on success.
642 static int ll_ioc_copy_start(struct super_block *sb, struct hsm_copy *copy)
644 struct ll_sb_info *sbi = ll_s2sbi(sb);
645 struct hsm_progress_kernel hpk;
649 /* Forge a hsm_progress based on data from copy. */
650 hpk.hpk_fid = copy->hc_hai.hai_fid;
651 hpk.hpk_cookie = copy->hc_hai.hai_cookie;
652 hpk.hpk_extent.offset = copy->hc_hai.hai_extent.offset;
653 hpk.hpk_extent.length = 0;
656 hpk.hpk_data_version = 0;
659 /* For archive request, we need to read the current file version. */
660 if (copy->hc_hai.hai_action == HSMA_ARCHIVE) {
662 __u64 data_version = 0;
664 /* Get inode for this fid */
665 inode = search_inode_for_lustre(sb, ©->hc_hai.hai_fid);
667 hpk.hpk_flags |= HP_FLAG_RETRY;
668 /* hpk_errval is >= 0 */
669 hpk.hpk_errval = -PTR_ERR(inode);
670 GOTO(progress, rc = PTR_ERR(inode));
673 /* Read current file data version */
674 rc = ll_data_version(inode, &data_version, LL_DV_RD_FLUSH);
677 CDEBUG(D_HSM, "Could not read file data version of "
678 DFID" (rc = %d). Archive request ("
679 LPX64") could not be done.\n",
680 PFID(©->hc_hai.hai_fid), rc,
681 copy->hc_hai.hai_cookie);
682 hpk.hpk_flags |= HP_FLAG_RETRY;
683 /* hpk_errval must be >= 0 */
684 hpk.hpk_errval = -rc;
688 /* Store it the hsm_copy for later copytool use.
689 * Always modified even if no lsm. */
690 copy->hc_data_version = data_version;
694 /* On error, the request should be considered as completed */
695 if (hpk.hpk_errval > 0)
696 hpk.hpk_flags |= HP_FLAG_COMPLETED;
697 rc = obd_iocontrol(LL_IOC_HSM_PROGRESS, sbi->ll_md_exp, sizeof(hpk),
704 * Generic handler to do any post-copy work.
706 * It will send the last hsm_progress update to coordinator to inform it
707 * that copy is finished and whether it was successful or not.
710 * - for ARCHIVE request, it will sample the file data version and compare it
711 * with the version saved in ll_ioc_copy_start(). If they do not match, copy
712 * will be considered as failed.
713 * - for RESTORE request, it will sample the file data version and send it to
714 * coordinator which is useful if the file was imported as 'released'.
716 * \return 0 on success.
718 static int ll_ioc_copy_end(struct super_block *sb, struct hsm_copy *copy)
720 struct ll_sb_info *sbi = ll_s2sbi(sb);
721 struct hsm_progress_kernel hpk;
725 /* If you modify the logic here, also check llapi_hsm_copy_end(). */
726 /* Take care: copy->hc_hai.hai_action, len, gid and data are not
727 * initialized if copy_end was called with copy == NULL.
730 /* Forge a hsm_progress based on data from copy. */
731 hpk.hpk_fid = copy->hc_hai.hai_fid;
732 hpk.hpk_cookie = copy->hc_hai.hai_cookie;
733 hpk.hpk_extent = copy->hc_hai.hai_extent;
734 hpk.hpk_flags = copy->hc_flags | HP_FLAG_COMPLETED;
735 hpk.hpk_errval = copy->hc_errval;
736 hpk.hpk_data_version = 0;
738 /* For archive request, we need to check the file data was not changed.
740 * For restore request, we need to send the file data version, this is
741 * useful when the file was created using hsm_import.
743 if (((copy->hc_hai.hai_action == HSMA_ARCHIVE) ||
744 (copy->hc_hai.hai_action == HSMA_RESTORE)) &&
745 (copy->hc_errval == 0)) {
747 __u64 data_version = 0;
749 /* Get lsm for this fid */
750 inode = search_inode_for_lustre(sb, ©->hc_hai.hai_fid);
752 hpk.hpk_flags |= HP_FLAG_RETRY;
753 /* hpk_errval must be >= 0 */
754 hpk.hpk_errval = -PTR_ERR(inode);
755 GOTO(progress, rc = PTR_ERR(inode));
758 rc = ll_data_version(inode, &data_version, LL_DV_RD_FLUSH);
761 CDEBUG(D_HSM, "Could not read file data version. "
762 "Request could not be confirmed.\n");
763 if (hpk.hpk_errval == 0)
764 hpk.hpk_errval = -rc;
768 /* Store it the hsm_copy for later copytool use.
769 * Always modified even if no lsm. */
770 hpk.hpk_data_version = data_version;
772 /* File could have been stripped during archiving, so we need
773 * to check anyway. */
774 if ((copy->hc_hai.hai_action == HSMA_ARCHIVE) &&
775 (copy->hc_data_version != data_version)) {
776 CDEBUG(D_HSM, "File data version mismatched. "
777 "File content was changed during archiving. "
778 DFID", start:"LPX64" current:"LPX64"\n",
779 PFID(©->hc_hai.hai_fid),
780 copy->hc_data_version, data_version);
781 /* File was changed, send error to cdt. Do not ask for
782 * retry because if a file is modified frequently,
783 * the cdt will loop on retried archive requests.
784 * The policy engine will ask for a new archive later
785 * when the file will not be modified for some tunable
787 /* we do not notify caller */
788 hpk.hpk_flags &= ~HP_FLAG_RETRY;
789 /* hpk_errval must be >= 0 */
790 hpk.hpk_errval = EBUSY;
796 rc = obd_iocontrol(LL_IOC_HSM_PROGRESS, sbi->ll_md_exp, sizeof(hpk),
803 static int copy_and_ioctl(int cmd, struct obd_export *exp,
804 const void __user *data, size_t size)
809 OBD_ALLOC(copy, size);
813 if (copy_from_user(copy, data, size)) {
818 rc = obd_iocontrol(cmd, exp, size, copy, NULL);
820 OBD_FREE(copy, size);
825 static int quotactl_ioctl(struct ll_sb_info *sbi, struct if_quotactl *qctl)
827 int cmd = qctl->qc_cmd;
828 int type = qctl->qc_type;
829 int id = qctl->qc_id;
830 int valid = qctl->qc_valid;
835 case LUSTRE_Q_INVALIDATE:
836 case LUSTRE_Q_FINVALIDATE:
841 if (!cfs_capable(CFS_CAP_SYS_ADMIN) ||
842 sbi->ll_flags & LL_SBI_RMT_CLIENT)
846 if (((type == USRQUOTA && current_euid() != id) ||
847 (type == GRPQUOTA && !in_egroup_p(id))) &&
848 (!cfs_capable(CFS_CAP_SYS_ADMIN) ||
849 sbi->ll_flags & LL_SBI_RMT_CLIENT))
855 CERROR("unsupported quotactl op: %#x\n", cmd);
859 if (valid != QC_GENERAL) {
860 if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
863 if (cmd == Q_GETINFO)
864 qctl->qc_cmd = Q_GETOINFO;
865 else if (cmd == Q_GETQUOTA)
866 qctl->qc_cmd = Q_GETOQUOTA;
872 rc = obd_iocontrol(OBD_IOC_QUOTACTL, sbi->ll_md_exp,
873 sizeof(*qctl), qctl, NULL);
876 rc = obd_iocontrol(OBD_IOC_QUOTACTL, sbi->ll_dt_exp,
877 sizeof(*qctl), qctl, NULL);
880 rc = obd_iocontrol(OBD_IOC_QUOTACTL, sbi->ll_md_exp,
881 sizeof(*qctl), qctl, NULL);
883 rc = obd_iocontrol(OBD_IOC_QUOTACTL,
885 sizeof(*qctl), qctl, NULL);
897 struct obd_quotactl *oqctl;
899 OBD_ALLOC_PTR(oqctl);
903 QCTL_COPY(oqctl, qctl);
904 rc = obd_quotactl(sbi->ll_md_exp, oqctl);
906 if (rc != -EALREADY && cmd == Q_QUOTAON) {
907 oqctl->qc_cmd = Q_QUOTAOFF;
908 obd_quotactl(sbi->ll_md_exp, oqctl);
913 /* If QIF_SPACE is not set, client should collect the
914 * space usage from OSSs by itself */
915 if (cmd == Q_GETQUOTA &&
916 !(oqctl->qc_dqblk.dqb_valid & QIF_SPACE) &&
917 !oqctl->qc_dqblk.dqb_curspace) {
918 struct obd_quotactl *oqctl_tmp;
920 OBD_ALLOC_PTR(oqctl_tmp);
921 if (oqctl_tmp == NULL)
922 GOTO(out, rc = -ENOMEM);
924 oqctl_tmp->qc_cmd = Q_GETOQUOTA;
925 oqctl_tmp->qc_id = oqctl->qc_id;
926 oqctl_tmp->qc_type = oqctl->qc_type;
928 /* collect space usage from OSTs */
929 oqctl_tmp->qc_dqblk.dqb_curspace = 0;
930 rc = obd_quotactl(sbi->ll_dt_exp, oqctl_tmp);
931 if (!rc || rc == -EREMOTEIO) {
932 oqctl->qc_dqblk.dqb_curspace =
933 oqctl_tmp->qc_dqblk.dqb_curspace;
934 oqctl->qc_dqblk.dqb_valid |= QIF_SPACE;
937 /* collect space & inode usage from MDTs */
938 oqctl_tmp->qc_dqblk.dqb_curspace = 0;
939 oqctl_tmp->qc_dqblk.dqb_curinodes = 0;
940 rc = obd_quotactl(sbi->ll_md_exp, oqctl_tmp);
941 if (!rc || rc == -EREMOTEIO) {
942 oqctl->qc_dqblk.dqb_curspace +=
943 oqctl_tmp->qc_dqblk.dqb_curspace;
944 oqctl->qc_dqblk.dqb_curinodes =
945 oqctl_tmp->qc_dqblk.dqb_curinodes;
946 oqctl->qc_dqblk.dqb_valid |= QIF_INODES;
948 oqctl->qc_dqblk.dqb_valid &= ~QIF_SPACE;
951 OBD_FREE_PTR(oqctl_tmp);
954 QCTL_COPY(qctl, oqctl);
962 ll_getname(const char __user *filename)
965 char *tmp = __getname();
968 return ERR_PTR(-ENOMEM);
970 len = strncpy_from_user(tmp, filename, PATH_MAX);
973 else if (len > PATH_MAX)
983 #define ll_putname(filename) __putname(filename)
985 static long ll_dir_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
987 struct inode *inode = file->f_dentry->d_inode;
988 struct ll_sb_info *sbi = ll_i2sbi(inode);
989 struct obd_ioctl_data *data;
993 CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p), cmd=%#x\n",
994 PFID(ll_inode2fid(inode)), inode, cmd);
996 /* asm-ppc{,64} declares TCGETS, et. al. as type 't' not 'T' */
997 if (_IOC_TYPE(cmd) == 'T' || _IOC_TYPE(cmd) == 't') /* tty ioctls */
1000 ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_IOCTL, 1);
1002 case FSFILT_IOC_GETFLAGS:
1003 case FSFILT_IOC_SETFLAGS:
1004 RETURN(ll_iocontrol(inode, file, cmd, arg));
1005 case FSFILT_IOC_GETVERSION_OLD:
1006 case FSFILT_IOC_GETVERSION:
1007 RETURN(put_user(inode->i_generation, (int *)arg));
1008 /* We need to special case any other ioctls we want to handle,
1009 * to send them to the MDS/OST as appropriate and to properly
1010 * network encode the arg field.
1011 case FSFILT_IOC_SETVERSION_OLD:
1012 case FSFILT_IOC_SETVERSION:
1014 case LL_IOC_GET_MDTIDX: {
1017 mdtidx = ll_get_mdt_idx(inode);
1021 if (put_user((int)mdtidx, (int*)arg))
1026 case IOC_MDC_LOOKUP: {
1027 struct ptlrpc_request *request = NULL;
1028 int namelen, len = 0;
1031 struct md_op_data *op_data;
1033 rc = obd_ioctl_getdata(&buf, &len, (void *)arg);
1038 filename = data->ioc_inlbuf1;
1039 namelen = strlen(filename);
1042 CDEBUG(D_INFO, "IOC_MDC_LOOKUP missing filename\n");
1043 GOTO(out_free, rc = -EINVAL);
1046 op_data = ll_prep_md_op_data(NULL, inode, NULL, filename, namelen,
1047 0, LUSTRE_OPC_ANY, NULL);
1048 if (IS_ERR(op_data))
1049 GOTO(out_free, rc = PTR_ERR(op_data));
1051 op_data->op_valid = OBD_MD_FLID;
1052 rc = md_getattr_name(sbi->ll_md_exp, op_data, &request);
1053 ll_finish_md_op_data(op_data);
1055 CDEBUG(D_INFO, "md_getattr_name: %d\n", rc);
1058 ptlrpc_req_finished(request);
1061 obd_ioctl_freedata(buf, len);
1064 case LL_IOC_LMV_SETSTRIPE: {
1065 struct lmv_user_md *lum;
1073 rc = obd_ioctl_getdata(&buf, &len, (void *)arg);
1078 if (data->ioc_inlbuf1 == NULL || data->ioc_inlbuf2 == NULL ||
1079 data->ioc_inllen1 == 0 || data->ioc_inllen2 == 0)
1080 GOTO(lmv_out_free, rc = -EINVAL);
1082 filename = data->ioc_inlbuf1;
1083 namelen = data->ioc_inllen1;
1086 CDEBUG(D_INFO, "IOC_MDC_LOOKUP missing filename\n");
1087 GOTO(lmv_out_free, rc = -EINVAL);
1089 lum = (struct lmv_user_md *)data->ioc_inlbuf2;
1090 lumlen = data->ioc_inllen2;
1092 if (lum->lum_magic != LMV_USER_MAGIC ||
1093 lumlen != sizeof(*lum)) {
1094 CERROR("%s: wrong lum magic %x or size %d: rc = %d\n",
1095 filename, lum->lum_magic, lumlen, -EFAULT);
1096 GOTO(lmv_out_free, rc = -EINVAL);
1100 * ll_dir_setdirstripe will be used to set dir stripe
1101 * mdc_create--->mdt_reint_create (with dirstripe)
1103 rc = ll_dir_setdirstripe(inode, lum, filename);
1105 obd_ioctl_freedata(buf, len);
1109 case LL_IOC_LMV_SET_DEFAULT_STRIPE: {
1110 struct lmv_user_md lum;
1111 struct lmv_user_md __user *ulump =
1112 (struct lmv_user_md __user *)arg;
1115 if (copy_from_user(&lum, ulump, sizeof(lum)))
1118 if (lum.lum_magic != LMV_USER_MAGIC)
1121 rc = ll_dir_setstripe(inode, (struct lov_user_md *)&lum, 0);
1125 case LL_IOC_LOV_SETSTRIPE: {
1126 struct lov_user_md_v3 lumv3;
1127 struct lov_user_md_v1 *lumv1 = (struct lov_user_md_v1 *)&lumv3;
1128 struct lov_user_md_v1 *lumv1p = (struct lov_user_md_v1 *)arg;
1129 struct lov_user_md_v3 *lumv3p = (struct lov_user_md_v3 *)arg;
1131 int set_default = 0;
1133 LASSERT(sizeof(lumv3) == sizeof(*lumv3p));
1134 LASSERT(sizeof(lumv3.lmm_objects[0]) ==
1135 sizeof(lumv3p->lmm_objects[0]));
1136 /* first try with v1 which is smaller than v3 */
1137 if (copy_from_user(lumv1, lumv1p, sizeof(*lumv1)))
1140 if ((lumv1->lmm_magic == LOV_USER_MAGIC_V3) ) {
1141 if (copy_from_user(&lumv3, lumv3p, sizeof(lumv3)))
1145 if (inode->i_sb->s_root == file->f_dentry)
1148 /* in v1 and v3 cases lumv1 points to data */
1149 rc = ll_dir_setstripe(inode, lumv1, set_default);
1153 case LL_IOC_LMV_GETSTRIPE: {
1154 struct lmv_user_md __user *ulmv =
1155 (struct lmv_user_md __user *)arg;
1156 struct lmv_user_md lum;
1157 struct ptlrpc_request *request = NULL;
1158 union lmv_mds_md *lmm = NULL;
1160 obd_valid valid = 0;
1161 struct lmv_user_md *tmp = NULL;
1168 if (copy_from_user(&lum, ulmv, sizeof(*ulmv)))
1171 /* lum_magic will indicate which stripe the ioctl will like
1172 * to get, LMV_MAGIC_V1 is for normal LMV stripe, LMV_USER_MAGIC
1173 * is for default LMV stripe */
1174 if (lum.lum_magic == LMV_MAGIC_V1)
1175 valid |= OBD_MD_MEA;
1176 else if (lum.lum_magic == LMV_USER_MAGIC)
1177 valid |= OBD_MD_DEFAULT_MEA;
1181 rc = ll_dir_getstripe(inode, (void **)&lmm, &lmmsize, &request,
1183 if (rc != 0 && rc != -ENODATA)
1184 GOTO(finish_req, rc);
1186 /* Get default LMV EA */
1187 if (lum.lum_magic == LMV_USER_MAGIC) {
1189 GOTO(finish_req, rc);
1191 if (lmmsize > sizeof(*ulmv))
1192 GOTO(finish_req, rc = -EINVAL);
1194 if (copy_to_user(ulmv, lmm, lmmsize))
1195 GOTO(finish_req, rc = -EFAULT);
1197 GOTO(finish_req, rc);
1200 /* Get normal LMV EA */
1201 if (rc == -ENODATA) {
1204 LASSERT(lmm != NULL);
1205 stripe_count = lmv_mds_md_stripe_count_get(lmm);
1208 lum_size = lmv_user_md_size(stripe_count, LMV_MAGIC_V1);
1209 OBD_ALLOC(tmp, lum_size);
1211 GOTO(finish_req, rc = -ENOMEM);
1213 tmp->lum_magic = LMV_MAGIC_V1;
1214 tmp->lum_stripe_count = 1;
1215 mdt_index = ll_get_mdt_idx(inode);
1217 GOTO(out_tmp, rc = -ENOMEM);
1218 tmp->lum_stripe_offset = mdt_index;
1219 tmp->lum_objects[0].lum_mds = mdt_index;
1220 tmp->lum_objects[0].lum_fid = *ll_inode2fid(inode);
1221 for (i = 1; i < stripe_count; i++) {
1222 struct lmv_mds_md_v1 *lmm1;
1224 lmm1 = &lmm->lmv_md_v1;
1225 mdt_index = ll_get_mdt_idx_by_fid(sbi,
1226 &lmm1->lmv_stripe_fids[i]);
1228 GOTO(out_tmp, rc = mdt_index);
1230 tmp->lum_objects[i].lum_mds = mdt_index;
1231 tmp->lum_objects[i].lum_fid = lmm1->lmv_stripe_fids[i];
1232 tmp->lum_stripe_count++;
1235 if (copy_to_user(ulmv, tmp, lum_size))
1236 GOTO(out_tmp, rc = -EFAULT);
1238 OBD_FREE(tmp, lum_size);
1240 ptlrpc_req_finished(request);
1244 case LL_IOC_REMOVE_ENTRY: {
1245 char *filename = NULL;
1249 /* Here is a little hack to avoid sending REINT_RMENTRY to
1250 * unsupported server, which might crash the server(LU-2730),
1251 * Because both LVB_TYPE and REINT_RMENTRY will be supported
1252 * on 2.4, we use OBD_CONNECT_LVB_TYPE to detect whether the
1253 * server will support REINT_RMENTRY XXX*/
1254 if (!(exp_connect_flags(sbi->ll_md_exp) & OBD_CONNECT_LVB_TYPE))
1257 filename = ll_getname((const char *)arg);
1258 if (IS_ERR(filename))
1259 RETURN(PTR_ERR(filename));
1261 namelen = strlen(filename);
1263 GOTO(out_rmdir, rc = -EINVAL);
1265 rc = ll_rmdir_entry(inode, filename, namelen);
1268 ll_putname(filename);
1271 case LL_IOC_LOV_SWAP_LAYOUTS:
1273 case LL_IOC_OBD_STATFS:
1274 RETURN(ll_obd_statfs(inode, (void *)arg));
1275 case LL_IOC_LOV_GETSTRIPE:
1276 case LL_IOC_MDC_GETINFO:
1277 case IOC_MDC_GETFILEINFO:
1278 case IOC_MDC_GETFILESTRIPE: {
1279 struct ptlrpc_request *request = NULL;
1280 struct lov_user_md *lump;
1281 struct lov_mds_md *lmm = NULL;
1282 struct mdt_body *body;
1283 char *filename = NULL;
1286 if (cmd == IOC_MDC_GETFILEINFO ||
1287 cmd == IOC_MDC_GETFILESTRIPE) {
1288 filename = ll_getname((const char *)arg);
1289 if (IS_ERR(filename))
1290 RETURN(PTR_ERR(filename));
1292 rc = ll_lov_getstripe_ea_info(inode, filename, &lmm,
1293 &lmmsize, &request);
1295 rc = ll_dir_getstripe(inode, (void **)&lmm, &lmmsize,
1300 body = req_capsule_server_get(&request->rq_pill,
1302 LASSERT(body != NULL);
1308 if (rc == -ENODATA && (cmd == IOC_MDC_GETFILEINFO ||
1309 cmd == LL_IOC_MDC_GETINFO))
1310 GOTO(skip_lmm, rc = 0);
1315 if (cmd == IOC_MDC_GETFILESTRIPE ||
1316 cmd == LL_IOC_LOV_GETSTRIPE) {
1317 lump = (struct lov_user_md *)arg;
1319 struct lov_user_mds_data *lmdp;
1320 lmdp = (struct lov_user_mds_data *)arg;
1321 lump = &lmdp->lmd_lmm;
1323 if (copy_to_user(lump, lmm, lmmsize)) {
1324 if (copy_to_user(lump, lmm, sizeof(*lump)))
1325 GOTO(out_req, rc = -EFAULT);
1329 if (cmd == IOC_MDC_GETFILEINFO || cmd == LL_IOC_MDC_GETINFO) {
1330 struct lov_user_mds_data *lmdp;
1333 st.st_dev = inode->i_sb->s_dev;
1334 st.st_mode = body->mode;
1335 st.st_nlink = body->nlink;
1336 st.st_uid = body->uid;
1337 st.st_gid = body->gid;
1338 st.st_rdev = body->rdev;
1339 st.st_size = body->size;
1340 st.st_blksize = PAGE_CACHE_SIZE;
1341 st.st_blocks = body->blocks;
1342 st.st_atime = body->atime;
1343 st.st_mtime = body->mtime;
1344 st.st_ctime = body->ctime;
1345 st.st_ino = inode->i_ino;
1347 lmdp = (struct lov_user_mds_data *)arg;
1348 if (copy_to_user(&lmdp->lmd_st, &st, sizeof(st)))
1349 GOTO(out_req, rc = -EFAULT);
1354 ptlrpc_req_finished(request);
1356 ll_putname(filename);
1359 case IOC_LOV_GETINFO: {
1360 struct lov_user_mds_data *lumd;
1361 struct lov_stripe_md *lsm;
1362 struct lov_user_md *lum;
1363 struct lov_mds_md *lmm;
1367 lumd = (struct lov_user_mds_data *)arg;
1368 lum = &lumd->lmd_lmm;
1370 rc = ll_get_max_mdsize(sbi, &lmmsize);
1374 OBD_ALLOC_LARGE(lmm, lmmsize);
1378 if (copy_from_user(lmm, lum, lmmsize))
1379 GOTO(free_lmm, rc = -EFAULT);
1381 switch (lmm->lmm_magic) {
1382 case LOV_USER_MAGIC_V1:
1383 if (LOV_USER_MAGIC_V1 == cpu_to_le32(LOV_USER_MAGIC_V1))
1385 /* swab objects first so that stripes num will be sane */
1386 lustre_swab_lov_user_md_objects(
1387 ((struct lov_user_md_v1 *)lmm)->lmm_objects,
1388 ((struct lov_user_md_v1 *)lmm)->lmm_stripe_count);
1389 lustre_swab_lov_user_md_v1((struct lov_user_md_v1 *)lmm);
1391 case LOV_USER_MAGIC_V3:
1392 if (LOV_USER_MAGIC_V3 == cpu_to_le32(LOV_USER_MAGIC_V3))
1394 /* swab objects first so that stripes num will be sane */
1395 lustre_swab_lov_user_md_objects(
1396 ((struct lov_user_md_v3 *)lmm)->lmm_objects,
1397 ((struct lov_user_md_v3 *)lmm)->lmm_stripe_count);
1398 lustre_swab_lov_user_md_v3((struct lov_user_md_v3 *)lmm);
1401 GOTO(free_lmm, rc = -EINVAL);
1404 rc = obd_unpackmd(sbi->ll_dt_exp, &lsm, lmm, lmmsize);
1406 GOTO(free_lmm, rc = -ENOMEM);
1408 /* Perform glimpse_size operation. */
1409 memset(&st, 0, sizeof(st));
1411 rc = ll_glimpse_ioctl(sbi, lsm, &st);
1415 if (copy_to_user(&lumd->lmd_st, &st, sizeof(st)))
1416 GOTO(free_lsm, rc = -EFAULT);
1420 obd_free_memmd(sbi->ll_dt_exp, &lsm);
1422 OBD_FREE_LARGE(lmm, lmmsize);
1425 case OBD_IOC_LLOG_CATINFO: {
1426 RETURN(-EOPNOTSUPP);
1428 case OBD_IOC_QUOTACHECK: {
1429 struct obd_quotactl *oqctl;
1432 if (!cfs_capable(CFS_CAP_SYS_ADMIN) ||
1433 sbi->ll_flags & LL_SBI_RMT_CLIENT)
1436 OBD_ALLOC_PTR(oqctl);
1439 oqctl->qc_type = arg;
1440 rc = obd_quotacheck(sbi->ll_md_exp, oqctl);
1442 CDEBUG(D_INFO, "md_quotacheck failed: rc %d\n", rc);
1446 rc = obd_quotacheck(sbi->ll_dt_exp, oqctl);
1448 CDEBUG(D_INFO, "obd_quotacheck failed: rc %d\n", rc);
1450 OBD_FREE_PTR(oqctl);
1453 case OBD_IOC_POLL_QUOTACHECK: {
1454 struct if_quotacheck *check;
1456 if (!cfs_capable(CFS_CAP_SYS_ADMIN) ||
1457 sbi->ll_flags & LL_SBI_RMT_CLIENT)
1460 OBD_ALLOC_PTR(check);
1464 rc = obd_iocontrol(cmd, sbi->ll_md_exp, 0, (void *)check,
1467 CDEBUG(D_QUOTA, "mdc ioctl %d failed: %d\n", cmd, rc);
1468 if (copy_to_user((void *)arg, check,
1470 CDEBUG(D_QUOTA, "copy_to_user failed\n");
1474 rc = obd_iocontrol(cmd, sbi->ll_dt_exp, 0, (void *)check,
1477 CDEBUG(D_QUOTA, "osc ioctl %d failed: %d\n", cmd, rc);
1478 if (copy_to_user((void *)arg, check,
1480 CDEBUG(D_QUOTA, "copy_to_user failed\n");
1484 OBD_FREE_PTR(check);
1487 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 7, 50, 0)
1488 case LL_IOC_QUOTACTL_18: {
1489 /* copy the old 1.x quota struct for internal use, then copy
1490 * back into old format struct. For 1.8 compatibility. */
1491 struct if_quotactl_18 *qctl_18;
1492 struct if_quotactl *qctl_20;
1494 OBD_ALLOC_PTR(qctl_18);
1498 OBD_ALLOC_PTR(qctl_20);
1500 GOTO(out_quotactl_18, rc = -ENOMEM);
1502 if (copy_from_user(qctl_18, (void *)arg, sizeof(*qctl_18)))
1503 GOTO(out_quotactl_20, rc = -ENOMEM);
1505 QCTL_COPY(qctl_20, qctl_18);
1506 qctl_20->qc_idx = 0;
1508 /* XXX: dqb_valid was borrowed as a flag to mark that
1509 * only mds quota is wanted */
1510 if (qctl_18->qc_cmd == Q_GETQUOTA &&
1511 qctl_18->qc_dqblk.dqb_valid) {
1512 qctl_20->qc_valid = QC_MDTIDX;
1513 qctl_20->qc_dqblk.dqb_valid = 0;
1514 } else if (qctl_18->obd_uuid.uuid[0] != '\0') {
1515 qctl_20->qc_valid = QC_UUID;
1516 qctl_20->obd_uuid = qctl_18->obd_uuid;
1518 qctl_20->qc_valid = QC_GENERAL;
1521 rc = quotactl_ioctl(sbi, qctl_20);
1524 QCTL_COPY(qctl_18, qctl_20);
1525 qctl_18->obd_uuid = qctl_20->obd_uuid;
1527 if (copy_to_user((void *)arg, qctl_18,
1533 OBD_FREE_PTR(qctl_20);
1535 OBD_FREE_PTR(qctl_18);
1539 #warning "remove old LL_IOC_QUOTACTL_18 compatibility code"
1540 #endif /* LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 7, 50, 0) */
1541 case LL_IOC_QUOTACTL: {
1542 struct if_quotactl *qctl;
1544 OBD_ALLOC_PTR(qctl);
1548 if (copy_from_user(qctl, (void *)arg, sizeof(*qctl)))
1549 GOTO(out_quotactl, rc = -EFAULT);
1551 rc = quotactl_ioctl(sbi, qctl);
1553 if (rc == 0 && copy_to_user((void *)arg, qctl, sizeof(*qctl)))
1560 case OBD_IOC_GETDTNAME:
1561 case OBD_IOC_GETMDNAME:
1562 RETURN(ll_get_obd_name(inode, cmd, arg));
1563 case LL_IOC_FLUSHCTX:
1564 RETURN(ll_flush_ctx(inode));
1565 #ifdef CONFIG_FS_POSIX_ACL
1566 case LL_IOC_RMTACL: {
1567 if (sbi->ll_flags & LL_SBI_RMT_CLIENT &&
1568 inode == inode->i_sb->s_root->d_inode) {
1569 struct ll_file_data *fd = LUSTRE_FPRIVATE(file);
1571 LASSERT(fd != NULL);
1572 rc = rct_add(&sbi->ll_rct, current_pid(), arg);
1574 fd->fd_flags |= LL_FILE_RMTACL;
1580 case LL_IOC_GETOBDCOUNT: {
1582 struct obd_export *exp;
1584 if (copy_from_user(&count, (int *)arg, sizeof(int)))
1587 /* get ost count when count is zero, get mdt count otherwise */
1588 exp = count ? sbi->ll_md_exp : sbi->ll_dt_exp;
1589 vallen = sizeof(count);
1590 rc = obd_get_info(NULL, exp, sizeof(KEY_TGT_COUNT),
1591 KEY_TGT_COUNT, &vallen, &count, NULL);
1593 CERROR("get target count failed: %d\n", rc);
1597 if (copy_to_user((int *)arg, &count, sizeof(int)))
1602 case LL_IOC_PATH2FID:
1603 if (copy_to_user((void *)arg, ll_inode2fid(inode),
1604 sizeof(struct lu_fid)))
1607 case LL_IOC_GET_CONNECT_FLAGS: {
1608 RETURN(obd_iocontrol(cmd, sbi->ll_md_exp, 0, NULL, (void*)arg));
1610 case OBD_IOC_CHANGELOG_SEND:
1611 case OBD_IOC_CHANGELOG_CLEAR:
1612 rc = copy_and_ioctl(cmd, sbi->ll_md_exp, (void *)arg,
1613 sizeof(struct ioc_changelog));
1615 case OBD_IOC_FID2PATH:
1616 RETURN(ll_fid2path(inode, (void *)arg));
1617 case LL_IOC_HSM_REQUEST: {
1618 struct hsm_user_request *hur;
1625 /* We don't know the true size yet; copy the fixed-size part */
1626 if (copy_from_user(hur, (void *)arg, sizeof(*hur))) {
1631 /* Compute the whole struct size */
1632 totalsize = hur_len(hur);
1635 /* Make sure the size is reasonable */
1636 if (totalsize >= MDS_MAXREQSIZE)
1639 OBD_ALLOC_LARGE(hur, totalsize);
1643 /* Copy the whole struct */
1644 if (copy_from_user(hur, (void *)arg, totalsize)) {
1645 OBD_FREE_LARGE(hur, totalsize);
1649 if (hur->hur_request.hr_action == HUA_RELEASE) {
1650 const struct lu_fid *fid;
1654 for (i = 0; i < hur->hur_request.hr_itemcount; i++) {
1655 fid = &hur->hur_user_item[i].hui_fid;
1656 f = search_inode_for_lustre(inode->i_sb, fid);
1662 rc = ll_hsm_release(f);
1668 rc = obd_iocontrol(cmd, ll_i2mdexp(inode), totalsize,
1672 OBD_FREE_LARGE(hur, totalsize);
1676 case LL_IOC_HSM_PROGRESS: {
1677 struct hsm_progress_kernel hpk;
1678 struct hsm_progress hp;
1680 if (copy_from_user(&hp, (void *)arg, sizeof(hp)))
1683 hpk.hpk_fid = hp.hp_fid;
1684 hpk.hpk_cookie = hp.hp_cookie;
1685 hpk.hpk_extent = hp.hp_extent;
1686 hpk.hpk_flags = hp.hp_flags;
1687 hpk.hpk_errval = hp.hp_errval;
1688 hpk.hpk_data_version = 0;
1690 /* File may not exist in Lustre; all progress
1691 * reported to Lustre root */
1692 rc = obd_iocontrol(cmd, sbi->ll_md_exp, sizeof(hpk), &hpk,
1696 case LL_IOC_HSM_CT_START:
1697 if (!cfs_capable(CFS_CAP_SYS_ADMIN))
1700 rc = copy_and_ioctl(cmd, sbi->ll_md_exp, (void *)arg,
1701 sizeof(struct lustre_kernelcomm));
1704 case LL_IOC_HSM_COPY_START: {
1705 struct hsm_copy *copy;
1708 OBD_ALLOC_PTR(copy);
1711 if (copy_from_user(copy, (char *)arg, sizeof(*copy))) {
1716 rc = ll_ioc_copy_start(inode->i_sb, copy);
1717 if (copy_to_user((char *)arg, copy, sizeof(*copy)))
1723 case LL_IOC_HSM_COPY_END: {
1724 struct hsm_copy *copy;
1727 OBD_ALLOC_PTR(copy);
1730 if (copy_from_user(copy, (char *)arg, sizeof(*copy))) {
1735 rc = ll_ioc_copy_end(inode->i_sb, copy);
1736 if (copy_to_user((char *)arg, copy, sizeof(*copy)))
1743 RETURN(obd_iocontrol(cmd, sbi->ll_dt_exp, 0, NULL,
1748 static loff_t ll_dir_seek(struct file *file, loff_t offset, int origin)
1750 struct inode *inode = file->f_mapping->host;
1751 struct ll_file_data *fd = LUSTRE_FPRIVATE(file);
1752 struct ll_sb_info *sbi = ll_i2sbi(inode);
1753 int api32 = ll_need_32bit_api(sbi);
1754 loff_t ret = -EINVAL;
1757 mutex_lock(&inode->i_mutex);
1762 offset += file->f_pos;
1768 offset += LL_DIR_END_OFF_32BIT;
1770 offset += LL_DIR_END_OFF;
1777 ((api32 && offset <= LL_DIR_END_OFF_32BIT) ||
1778 (!api32 && offset <= LL_DIR_END_OFF))) {
1779 if (offset != file->f_pos) {
1780 if ((api32 && offset == LL_DIR_END_OFF_32BIT) ||
1781 (!api32 && offset == LL_DIR_END_OFF))
1782 fd->lfd_pos = MDS_DIR_END_OFF;
1783 else if (api32 && sbi->ll_flags & LL_SBI_64BIT_HASH)
1784 fd->lfd_pos = offset << 32;
1786 fd->lfd_pos = offset;
1787 file->f_pos = offset;
1788 file->f_version = 0;
1795 mutex_unlock(&inode->i_mutex);
1799 int ll_dir_open(struct inode *inode, struct file *file)
1802 RETURN(ll_file_open(inode, file));
1805 int ll_dir_release(struct inode *inode, struct file *file)
1808 RETURN(ll_file_release(inode, file));
1811 struct file_operations ll_dir_operations = {
1812 .llseek = ll_dir_seek,
1813 .open = ll_dir_open,
1814 .release = ll_dir_release,
1815 .read = generic_read_dir,
1816 .readdir = ll_readdir,
1817 .unlocked_ioctl = ll_dir_ioctl,