1 /* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
2 * vim:expandtab:shiftwidth=8:tabstop=8:
4 * Copyright (C) 1992, 1993, 1994, 1995
5 * Remy Card (card@masi.ibp.fr)
6 * Laboratoire MASI - Institut Blaise Pascal
7 * Universite Pierre et Marie Curie (Paris VI)
11 * linux/fs/minix/dir.c
14 * Copyright (C) 1991, 1992 Linus Torvalds
16 * ext2 directory handling functions
18 * Big-endian to little-endian byte-swapping/bitmaps by
19 * David S. Miller (davem@caip.rutgers.edu), 1995
21 * All code that works with directory layout had been switched to pagecache
24 * Adapted for Lustre Light
25 * Copyright (C) 2002-2003, Cluster File Systems, Inc.
30 #include <linux/ext2_fs.h>
31 #include <linux/pagemap.h>
33 #include <linux/version.h>
34 #include <linux/smp_lock.h>
35 #include <asm/uaccess.h>
36 #include <linux/file.h>
37 #include <linux/kmod.h>
38 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,5,0))
39 # include <linux/locks.h> // for wait_on_buffer
41 # include <linux/buffer_head.h> // for wait_on_buffer
44 #define DEBUG_SUBSYSTEM S_LLITE
46 #include <linux/obd_support.h>
47 #include <linux/obd_class.h>
48 #include <linux/lustre_lib.h>
49 #include <linux/lustre_idl.h>
50 #include <linux/lustre_mds.h>
51 #include <linux/lustre_lite.h>
52 #include <linux/lustre_dlm.h>
53 #include "llite_internal.h"
55 typedef struct ext2_dir_entry_2 ext2_dirent;
57 #define PageChecked(page) test_bit(PG_checked, &(page)->flags)
58 #define SetPageChecked(page) set_bit(PG_checked, &(page)->flags)
60 /* returns the page unlocked, but with a reference */
61 static int ll_dir_readpage(struct file *file, struct page *page)
63 struct inode *inode = page->mapping->host;
64 struct ll_fid mdc_fid;
66 struct ptlrpc_request *request;
67 struct mds_body *body;
71 CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p)\n", inode->i_ino,
72 inode->i_generation, inode);
74 mdc_pack_fid(&mdc_fid, inode->i_ino, inode->i_generation, S_IFDIR);
75 mdc_fid.mds = ll_i2info(inode)->lli_mds;
76 offset = page->index << PAGE_SHIFT;
77 rc = md_readpage(ll_i2sbi(inode)->ll_mdc_exp, &mdc_fid,
78 offset, page, &request);
80 body = lustre_msg_buf(request->rq_repmsg, 0, sizeof (*body));
81 LASSERT (body != NULL); /* checked by md_readpage() */
82 LASSERT_REPSWABBED (request, 0); /* swabbed by md_readpage() */
84 #warning "FIXME ASAP!"
85 //inode->i_size = body->size;
86 SetPageUptodate(page);
88 ptlrpc_req_finished(request);
95 struct address_space_operations ll_dir_aops = {
96 .readpage = ll_dir_readpage,
100 * ext2 uses block-sized chunks. Arguably, sector-sized ones would be
101 * more robust, but we have what we have
103 static inline unsigned ext2_chunk_size(struct inode *inode)
105 return inode->i_sb->s_blocksize;
108 static inline void ext2_put_page(struct page *page)
111 page_cache_release(page);
114 static inline unsigned long dir_pages(struct inode *inode)
116 return (inode->i_size+PAGE_CACHE_SIZE-1)>>PAGE_CACHE_SHIFT;
120 static void ext2_check_page(struct page *page)
122 struct inode *dir = page->mapping->host;
123 unsigned chunk_size = ext2_chunk_size(dir);
124 char *kaddr = page_address(page);
125 // u32 max_inumber = le32_to_cpu(sb->u.ext2_sb.s_es->s_inodes_count);
126 unsigned offs, rec_len;
127 unsigned limit = PAGE_CACHE_SIZE;
131 if ((dir->i_size >> PAGE_CACHE_SHIFT) == page->index) {
132 limit = dir->i_size & ~PAGE_CACHE_MASK;
133 if (limit & (chunk_size - 1)) {
134 CERROR("limit %d dir size %lld index %ld\n",
135 limit, dir->i_size, page->index);
138 for (offs = limit; offs<PAGE_CACHE_SIZE; offs += chunk_size) {
139 ext2_dirent *p = (ext2_dirent*)(kaddr + offs);
140 p->rec_len = cpu_to_le16(chunk_size);
147 for (offs = 0; offs <= limit - EXT2_DIR_REC_LEN(1); offs += rec_len) {
148 p = (ext2_dirent *)(kaddr + offs);
149 rec_len = le16_to_cpu(p->rec_len);
151 if (rec_len < EXT2_DIR_REC_LEN(1))
155 if (rec_len < EXT2_DIR_REC_LEN(p->name_len))
157 if (((offs + rec_len - 1) ^ offs) & ~(chunk_size-1))
159 // if (le32_to_cpu(p->inode) > max_inumber)
165 SetPageChecked(page);
168 /* Too bad, we had an error */
171 CERROR("ext2_check_page"
172 "size of directory #%lu is not a multiple of chunk size\n",
177 error = "rec_len is smaller than minimal";
180 error = "unaligned directory entry";
183 error = "rec_len is too small for name_len";
186 error = "directory entry across blocks";
189 // error = "inode out of bounds";
191 CERROR("ext2_check_page: bad entry in directory #%lu: %s - "
192 "offset=%lu+%u, inode=%lu, rec_len=%d, name_len=%d",
193 dir->i_ino, error, (page->index<<PAGE_CACHE_SHIFT), offs,
194 (unsigned long) le32_to_cpu(p->inode),
195 rec_len, p->name_len);
198 p = (ext2_dirent *)(kaddr + offs);
199 CERROR("ext2_check_page"
200 "entry in directory #%lu spans the page boundary"
201 "offset=%lu, inode=%lu",
202 dir->i_ino, (page->index<<PAGE_CACHE_SHIFT)+offs,
203 (unsigned long) le32_to_cpu(p->inode));
205 SetPageChecked(page);
209 static struct page *ll_get_dir_page(struct inode *dir, unsigned long n)
211 struct ldlm_res_id res_id =
212 { .name = { dir->i_ino, (__u64)dir->i_generation} };
213 struct lustre_handle lockh;
214 struct obd_device *obddev = class_exp2obd(ll_i2sbi(dir)->ll_mdc_exp);
215 struct address_space *mapping = dir->i_mapping;
217 ldlm_policy_data_t policy = { .l_inodebits = { MDS_INODELOCK_UPDATE } };
220 obddev = md_get_real_obd(ll_i2sbi(dir)->ll_mdc_exp, NULL, 0);
221 rc = ldlm_lock_match(obddev->obd_namespace, LDLM_FL_BLOCK_GRANTED,
222 &res_id, LDLM_IBITS, &policy, LCK_PR, &lockh);
224 struct lookup_intent it = { .it_op = IT_READDIR };
225 struct ptlrpc_request *request;
226 struct mdc_op_data data;
228 ll_prepare_mdc_op_data(&data, dir, NULL, NULL, 0, 0);
230 rc = md_enqueue(ll_i2sbi(dir)->ll_mdc_exp, LDLM_IBITS, &it,
231 LCK_PR, &data, &lockh, NULL, 0,
232 ldlm_completion_ast, ll_mdc_blocking_ast, dir);
234 request = (struct ptlrpc_request *)it.d.lustre.it_data;
236 ptlrpc_req_finished(request);
238 CERROR("lock enqueue: rc: %d\n", rc);
242 ldlm_lock_dump_handle(D_OTHER, &lockh);
244 page = read_cache_page(mapping, n,
245 (filler_t*)mapping->a_ops->readpage, NULL);
249 if (!PageUptodate(page))
251 if (!PageChecked(page))
252 ext2_check_page(page);
258 ldlm_lock_decref(&lockh, LCK_PR);
263 page = ERR_PTR(-EIO);
268 * p is at least 6 bytes before the end of page
270 static inline ext2_dirent *ext2_next_entry(ext2_dirent *p)
272 return (ext2_dirent *)((char*)p + le16_to_cpu(p->rec_len));
275 static inline unsigned
276 ext2_validate_entry(char *base, unsigned offset, unsigned mask)
278 ext2_dirent *de = (ext2_dirent*)(base + offset);
279 ext2_dirent *p = (ext2_dirent*)(base + (offset&mask));
280 while ((char*)p < (char*)de)
281 p = ext2_next_entry(p);
282 return (char *)p - base;
285 static unsigned char ext2_filetype_table[EXT2_FT_MAX] = {
286 [EXT2_FT_UNKNOWN] DT_UNKNOWN,
287 [EXT2_FT_REG_FILE] DT_REG,
288 [EXT2_FT_DIR] DT_DIR,
289 [EXT2_FT_CHRDEV] DT_CHR,
290 [EXT2_FT_BLKDEV] DT_BLK,
291 [EXT2_FT_FIFO] DT_FIFO,
292 [EXT2_FT_SOCK] DT_SOCK,
293 [EXT2_FT_SYMLINK] DT_LNK,
297 int ll_readdir(struct file * filp, void * dirent, filldir_t filldir)
299 struct inode *inode = filp->f_dentry->d_inode;
300 loff_t pos = filp->f_pos;
301 // XXX struct super_block *sb = inode->i_sb;
302 unsigned offset = pos & ~PAGE_CACHE_MASK;
303 unsigned long n = pos >> PAGE_CACHE_SHIFT;
304 unsigned long npages = dir_pages(inode);
305 unsigned chunk_mask = ~(ext2_chunk_size(inode)-1);
306 unsigned char *types = NULL;
307 int need_revalidate = (filp->f_version != inode->i_version);
311 CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p) pos %llu/%llu\n",
312 inode->i_ino, inode->i_generation, inode, pos, inode->i_size);
314 if (pos > inode->i_size - EXT2_DIR_REC_LEN(1))
317 types = ext2_filetype_table;
319 for ( ; n < npages; n++, offset = 0) {
324 CDEBUG(D_EXT2,"read %lu of dir %lu/%u page %lu/%lu size %llu\n",
325 PAGE_CACHE_SIZE, inode->i_ino, inode->i_generation,
326 n, npages, inode->i_size);
327 page = ll_get_dir_page(inode, n);
329 /* size might have been updated by md_readpage() */
330 npages = dir_pages(inode);
334 CERROR("error reading dir %lu/%u page %lu: rc %d\n",
335 inode->i_ino, inode->i_generation, n, rc);
339 kaddr = page_address(page);
340 if (need_revalidate) {
341 offset = ext2_validate_entry(kaddr, offset, chunk_mask);
344 de = (ext2_dirent *)(kaddr+offset);
345 limit = kaddr + PAGE_CACHE_SIZE - EXT2_DIR_REC_LEN(1);
346 for ( ;(char*)de <= limit; de = ext2_next_entry(de)) {
349 unsigned char d_type = DT_UNKNOWN;
352 rc = 0; /* no error if we return something */
353 if (types && de->file_type < EXT2_FT_MAX)
354 d_type = types[de->file_type];
356 offset = (char *)de - kaddr;
357 over = filldir(dirent, de->name, de->name_len,
358 (n<<PAGE_CACHE_SHIFT) | offset,
359 le32_to_cpu(de->inode), d_type);
370 filp->f_pos = (n << PAGE_CACHE_SHIFT) | offset;
371 filp->f_version = inode->i_version;
376 static int ll_mkdir_stripe(struct inode *inode, unsigned long arg)
379 struct ll_user_mkdir_stripe lums;
380 struct ptlrpc_request *request = NULL;
381 struct ll_sb_info *sbi = ll_i2sbi(inode);
382 struct mdc_op_data op_data;
388 if (copy_from_user(&lums, (void *)arg, sizeof(lums)))
391 if (lums.lums_namelen <= 0)
393 OBD_ALLOC(name, lums.lums_namelen);
397 if (copy_from_user(name, lums.lums_name, lums.lums_namelen))
398 GOTO(out, err=-EFAULT);
400 CDEBUG(D_VFSTRACE, "ioctl Op:name=%s,dir=%lu/%u(%p)\n",
401 name, inode->i_ino, inode->i_generation, inode);
402 nstripes = lums.lums_nstripes;
404 mode = lums.lums_mode;
405 mode = (mode & (S_IRWXUGO|S_ISVTX) & ~current->fs->umask) | S_IFDIR;
406 ll_prepare_mdc_op_data(&op_data, inode, NULL, name,lums.lums_namelen,0);
407 err = md_create(sbi->ll_mdc_exp, &op_data, &nstripes, sizeof(nstripes),
408 mode, current->fsuid, current->fsgid, 0, &request);
409 ptlrpc_req_finished(request);
412 OBD_FREE(name, lums.lums_namelen);
416 static int ll_dir_ioctl(struct inode *inode, struct file *file,
417 unsigned int cmd, unsigned long arg)
419 struct ll_sb_info *sbi = ll_i2sbi(inode);
420 struct obd_ioctl_data *data;
423 CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p), cmd=%#x\n",
424 inode->i_ino, inode->i_generation, inode, cmd);
426 if (_IOC_TYPE(cmd) == 'T') /* tty ioctls */
429 lprocfs_counter_incr(ll_i2sbi(inode)->ll_stats, LPROC_LL_IOCTL);
431 case EXT3_IOC_GETFLAGS:
432 case EXT3_IOC_SETFLAGS:
433 RETURN(ll_iocontrol(inode, file, cmd, arg));
434 case IOC_MDC_LOOKUP: {
435 struct ptlrpc_request *request = NULL;
439 int namelen, rc, len = 0;
442 rc = obd_ioctl_getdata(&buf, &len, (void *)arg);
447 filename = data->ioc_inlbuf1;
448 namelen = data->ioc_inllen1;
451 CDEBUG(D_INFO, "IOC_MDC_LOOKUP missing filename\n");
452 GOTO(out, rc = -EINVAL);
456 ll_inode2fid(&fid, inode);
457 rc = md_getattr_name(sbi->ll_mdc_exp, &fid,
458 filename, namelen, valid, 0, &request);
460 CDEBUG(D_INFO, "md_getattr_name: %d\n", rc);
464 ptlrpc_req_finished(request);
468 obd_ioctl_freedata(buf, len);
471 case LL_IOC_MDC_MKDIRSTRIPE:
472 RETURN(ll_mkdir_stripe(inode, arg));
473 case IOC_MDC_FINISH_GNS:
474 RETURN(ll_finish_gns(sbi));
475 case LL_IOC_LOV_SETSTRIPE: {
476 struct ptlrpc_request *request = NULL;
477 struct mdc_op_data op_data;
478 struct iattr attr = { 0 };
479 struct lov_user_md lum, *lump = (struct lov_user_md *)arg;
482 ll_prepare_mdc_op_data(&op_data, inode, NULL, NULL, 0, 0);
484 LASSERT(sizeof(lum) == sizeof(*lump));
485 LASSERT(sizeof(lum.lmm_objects[0]) ==
486 sizeof(lump->lmm_objects[0]));
487 rc = copy_from_user(&lum, lump, sizeof(lum));
491 if (lum.lmm_magic != LOV_USER_MAGIC)
494 rc = md_setattr(sbi->ll_mdc_exp, &op_data,
495 &attr, &lum, sizeof(lum), NULL, 0, &request);
497 ptlrpc_req_finished(request);
498 if (rc != -EPERM && rc != -EACCES)
499 CERROR("md_setattr fails: rc = %d\n", rc);
502 ptlrpc_req_finished(request);
506 case LL_IOC_LOV_GETSTRIPE: {
507 struct ptlrpc_request *request = NULL;
508 struct lov_user_md *lump = (struct lov_user_md *)arg;
509 struct lov_mds_md *lmm;
511 struct mds_body *body;
512 unsigned long valid = 0;
515 valid |= OBD_MD_FLDIREA;
517 ll_inode2fid(&fid, inode);
518 rc = md_getattr(sbi->ll_mdc_exp, &fid, valid,
519 obd_size_diskmd(sbi->ll_osc_exp, NULL),
522 CDEBUG(D_INFO, "md_getattr failed: rc = %d\n", rc);
526 body = lustre_msg_buf(request->rq_repmsg, 0, sizeof(*body));
527 LASSERT(body != NULL); /* checked by md_getattr_name */
528 LASSERT_REPSWABBED(request, 0);/* swabbed by md_getattr_name */
530 lmmsize = body->eadatasize;
532 GOTO(out_get, rc = -ENODATA);
534 lmm = lustre_msg_buf(request->rq_repmsg, 1, lmmsize);
535 LASSERT(lmm != NULL);
536 LASSERT_REPSWABBED(request, 1);
537 rc = copy_to_user(lump, lmm, lmmsize);
539 GOTO(out_get, rc = -EFAULT);
543 ptlrpc_req_finished(request);
546 case IOC_MDC_GETSTRIPE: {
547 struct ptlrpc_request *request = NULL;
549 struct mds_body *body;
550 struct lov_user_md *lump = (struct lov_user_md *)arg;
551 struct lov_mds_md *lmm;
555 filename = getname((const char *)arg);
556 if (IS_ERR(filename))
557 RETURN(PTR_ERR(filename));
559 ll_inode2fid(&fid, inode);
560 rc = md_getattr_name(sbi->ll_mdc_exp, &fid, filename,
561 strlen(filename) + 1, OBD_MD_FLEASIZE,
562 obd_size_diskmd(sbi->ll_osc_exp, NULL),
565 CDEBUG(D_INFO, "md_getattr_name failed on %s: rc %d\n",
570 body = lustre_msg_buf(request->rq_repmsg, 0, sizeof (*body));
571 LASSERT(body != NULL); /* checked by md_getattr_name */
572 LASSERT_REPSWABBED(request, 0);/* swabbed by md_getattr_name */
574 lmmsize = body->eadatasize;
576 if (!(body->valid & OBD_MD_FLEASIZE) || lmmsize == 0)
577 GOTO(out_req, rc = -ENODATA);
580 GOTO(out_req, rc = -EFBIG);
582 lmm = lustre_msg_buf(request->rq_repmsg, 1, lmmsize);
583 LASSERT(lmm != NULL);
584 LASSERT_REPSWABBED(request, 1);
586 rc = copy_to_user(lump, lmm, lmmsize);
588 GOTO(out_req, rc = -EFAULT);
592 ptlrpc_req_finished(request);
598 struct ptlrpc_request *req = NULL;
601 struct client_obd *cli;
602 struct obd_device *obd;
604 rc = obd_ioctl_getdata(&buf, &len, (void *)arg);
609 obd = class_name2obd(data->ioc_inlbuf1);
612 GOTO(out_ping, rc = -ENODEV);
614 if (!obd->obd_attached) {
615 CERROR("Device %d not attached\n", obd->obd_minor);
616 GOTO(out_ping, rc = -ENODEV);
618 if (!obd->obd_set_up) {
619 CERROR("Device %d still not setup\n", obd->obd_minor);
620 GOTO(out_ping, rc = -ENODEV);
623 req = ptlrpc_prep_req(cli->cl_import, LUSTRE_OBD_VERSION,
624 OBD_PING, 0, NULL, NULL);
626 GOTO(out_ping, rc = -ENOMEM);
628 req->rq_replen = lustre_msg_size(0, NULL);
629 req->rq_send_state = LUSTRE_IMP_FULL;
631 rc = ptlrpc_queue_wait(req);
633 ptlrpc_req_finished(req);
635 obd_ioctl_freedata(buf, len);
638 case OBD_IOC_LLOG_CATINFO: {
639 struct ptlrpc_request *req = NULL;
645 rc = obd_ioctl_getdata(&buf, &len, (void *)arg);
650 if (!data->ioc_inlbuf1) {
651 obd_ioctl_freedata(buf, len);
655 lens[0] = data->ioc_inllen1;
656 bufs[0] = data->ioc_inlbuf1;
657 if (data->ioc_inllen2) {
658 lens[1] = data->ioc_inllen2;
659 bufs[1] = data->ioc_inlbuf2;
664 size = data->ioc_plen1;
665 req = ptlrpc_prep_req(sbi2mdc(sbi)->cl_import,
666 LUSTRE_LOG_VERSION, LLOG_CATINFO,
669 GOTO(out_catinfo, rc = -ENOMEM);
671 req->rq_replen = lustre_msg_size(1, &size);
673 rc = ptlrpc_queue_wait(req);
674 str = lustre_msg_string(req->rq_repmsg, 0, data->ioc_plen1);
676 rc = copy_to_user(data->ioc_pbuf1, str,
678 ptlrpc_req_finished(req);
680 obd_ioctl_freedata(buf, len);
684 return obd_iocontrol(cmd, sbi->ll_osc_exp,0,NULL,(void *)arg);
688 int ll_dir_open(struct inode *inode, struct file *file)
691 RETURN(ll_file_open(inode, file));
694 int ll_dir_release(struct inode *inode, struct file *file)
697 RETURN(ll_file_release(inode, file));
700 struct file_operations ll_dir_operations = {
702 .release = ll_dir_release,
703 .read = generic_read_dir,
704 .readdir = ll_readdir,
705 .ioctl = ll_dir_ioctl