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 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,5,0))
37 # include <linux/locks.h> // for wait_on_buffer
39 # include <linux/buffer_head.h> // for wait_on_buffer
42 #define DEBUG_SUBSYSTEM S_LLITE
44 #include <linux/obd_support.h>
45 #include <linux/obd_class.h>
46 #include <linux/lustre_lib.h>
47 #include <linux/lustre_idl.h>
48 #include <linux/lustre_mds.h>
49 #include <linux/lustre_lite.h>
50 #include <linux/lustre_dlm.h>
51 #include "llite_internal.h"
53 typedef struct ext2_dir_entry_2 ext2_dirent;
55 #define PageChecked(page) test_bit(PG_checked, &(page)->flags)
56 #define SetPageChecked(page) set_bit(PG_checked, &(page)->flags)
58 /* returns the page unlocked, but with a reference */
59 static int ll_dir_readpage(struct file *file, struct page *page)
61 struct inode *inode = page->mapping->host;
62 struct ll_fid mdc_fid;
64 struct ptlrpc_request *request;
65 struct mds_body *body;
69 CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p)\n", inode->i_ino,
70 inode->i_generation, inode);
72 mdc_pack_fid(&mdc_fid, inode->i_ino, inode->i_generation, S_IFDIR);
73 mdc_fid.mds = ll_i2info(inode)->lli_mds;
74 offset = page->index << PAGE_SHIFT;
75 rc = md_readpage(ll_i2sbi(inode)->ll_mdc_exp, &mdc_fid,
76 offset, page, &request);
78 body = lustre_msg_buf(request->rq_repmsg, 0, sizeof (*body));
79 LASSERT (body != NULL); /* checked by md_readpage() */
80 LASSERT_REPSWABBED (request, 0);/*swabbed by md_readpage()*/
82 #warning "FIXME ASAP!"
83 //inode->i_size = body->size;
84 SetPageUptodate(page);
86 ptlrpc_req_finished(request);
93 struct address_space_operations ll_dir_aops = {
94 readpage: ll_dir_readpage,
98 * ext2 uses block-sized chunks. Arguably, sector-sized ones would be
99 * more robust, but we have what we have
101 static inline unsigned ext2_chunk_size(struct inode *inode)
103 return inode->i_sb->s_blocksize;
106 static inline void ext2_put_page(struct page *page)
109 page_cache_release(page);
112 static inline unsigned long dir_pages(struct inode *inode)
114 return (inode->i_size+PAGE_CACHE_SIZE-1)>>PAGE_CACHE_SHIFT;
118 static void ext2_check_page(struct page *page)
120 struct inode *dir = page->mapping->host;
121 unsigned chunk_size = ext2_chunk_size(dir);
122 char *kaddr = page_address(page);
123 // u32 max_inumber = le32_to_cpu(sb->u.ext2_sb.s_es->s_inodes_count);
124 unsigned offs, rec_len;
125 unsigned limit = PAGE_CACHE_SIZE;
129 if ((dir->i_size >> PAGE_CACHE_SHIFT) == page->index) {
130 limit = dir->i_size & ~PAGE_CACHE_MASK;
131 if (limit & (chunk_size - 1)) {
132 CERROR("limit %d dir size %lld index %ld\n",
133 limit, dir->i_size, page->index);
136 for (offs = limit; offs<PAGE_CACHE_SIZE; offs += chunk_size) {
137 ext2_dirent *p = (ext2_dirent*)(kaddr + offs);
138 p->rec_len = cpu_to_le16(chunk_size);
145 for (offs = 0; offs <= limit - EXT2_DIR_REC_LEN(1); offs += rec_len) {
146 p = (ext2_dirent *)(kaddr + offs);
147 rec_len = le16_to_cpu(p->rec_len);
149 if (rec_len < EXT2_DIR_REC_LEN(1))
153 if (rec_len < EXT2_DIR_REC_LEN(p->name_len))
155 if (((offs + rec_len - 1) ^ offs) & ~(chunk_size-1))
157 // if (le32_to_cpu(p->inode) > max_inumber)
163 SetPageChecked(page);
166 /* Too bad, we had an error */
169 CERROR("ext2_check_page"
170 "size of directory #%lu is not a multiple of chunk size\n",
175 error = "rec_len is smaller than minimal";
178 error = "unaligned directory entry";
181 error = "rec_len is too small for name_len";
184 error = "directory entry across blocks";
187 // error = "inode out of bounds";
189 CERROR("ext2_check_page: bad entry in directory #%lu: %s - "
190 "offset=%lu+%u, inode=%lu, rec_len=%d, name_len=%d",
191 dir->i_ino, error, (page->index<<PAGE_CACHE_SHIFT), offs,
192 (unsigned long) le32_to_cpu(p->inode),
193 rec_len, p->name_len);
196 p = (ext2_dirent *)(kaddr + offs);
197 CERROR("ext2_check_page"
198 "entry in directory #%lu spans the page boundary"
199 "offset=%lu, inode=%lu",
200 dir->i_ino, (page->index<<PAGE_CACHE_SHIFT)+offs,
201 (unsigned long) le32_to_cpu(p->inode));
203 SetPageChecked(page);
208 static struct page *ll_get_dir_page(struct inode *dir, unsigned long n)
210 struct ldlm_res_id res_id =
211 { .name = { dir->i_ino, (__u64)dir->i_generation} };
212 struct lustre_handle lockh;
213 struct obd_device *obddev = class_exp2obd(ll_i2sbi(dir)->ll_mdc_exp);
214 struct address_space *mapping = dir->i_mapping;
216 ldlm_policy_data_t policy = { .l_inodebits = { MDS_INODELOCK_UPDATE } };
219 obddev = md_get_real_obd(ll_i2sbi(dir)->ll_mdc_exp, NULL, 0);
220 rc = ldlm_lock_match(obddev->obd_namespace, LDLM_FL_BLOCK_GRANTED,
221 &res_id, LDLM_IBITS, &policy, LCK_PR, &lockh);
223 struct lookup_intent it = { .it_op = IT_READDIR };
224 struct ptlrpc_request *request;
225 struct mdc_op_data data;
227 ll_prepare_mdc_op_data(&data, dir, NULL, NULL, 0, 0);
229 rc = md_enqueue(ll_i2sbi(dir)->ll_mdc_exp, LDLM_IBITS, &it,
230 LCK_PR, &data, &lockh, NULL, 0,
231 ldlm_completion_ast, ll_mdc_blocking_ast, dir);
233 request = (struct ptlrpc_request *)it.d.lustre.it_data;
235 ptlrpc_req_finished(request);
237 CERROR("lock enqueue: rc: %d\n", rc);
241 ldlm_lock_dump_handle(D_OTHER, &lockh);
243 page = read_cache_page(mapping, n,
244 (filler_t*)mapping->a_ops->readpage, NULL);
248 if (!PageUptodate(page))
250 if (!PageChecked(page))
251 ext2_check_page(page);
257 ldlm_lock_decref(&lockh, LCK_PR);
262 page = ERR_PTR(-EIO);
267 * p is at least 6 bytes before the end of page
269 static inline ext2_dirent *ext2_next_entry(ext2_dirent *p)
271 return (ext2_dirent *)((char*)p + le16_to_cpu(p->rec_len));
274 static inline unsigned
275 ext2_validate_entry(char *base, unsigned offset, unsigned mask)
277 ext2_dirent *de = (ext2_dirent*)(base + offset);
278 ext2_dirent *p = (ext2_dirent*)(base + (offset&mask));
279 while ((char*)p < (char*)de)
280 p = ext2_next_entry(p);
281 return (char *)p - base;
284 static unsigned char ext2_filetype_table[EXT2_FT_MAX] = {
285 [EXT2_FT_UNKNOWN] DT_UNKNOWN,
286 [EXT2_FT_REG_FILE] DT_REG,
287 [EXT2_FT_DIR] DT_DIR,
288 [EXT2_FT_CHRDEV] DT_CHR,
289 [EXT2_FT_BLKDEV] DT_BLK,
290 [EXT2_FT_FIFO] DT_FIFO,
291 [EXT2_FT_SOCK] DT_SOCK,
292 [EXT2_FT_SYMLINK] DT_LNK,
296 int ll_readdir(struct file * filp, void * dirent, filldir_t filldir)
298 struct inode *inode = filp->f_dentry->d_inode;
299 loff_t pos = filp->f_pos;
300 // XXX struct super_block *sb = inode->i_sb;
301 unsigned offset = pos & ~PAGE_CACHE_MASK;
302 unsigned long n = pos >> PAGE_CACHE_SHIFT;
303 unsigned long npages = dir_pages(inode);
304 unsigned chunk_mask = ~(ext2_chunk_size(inode)-1);
305 unsigned char *types = NULL;
306 int need_revalidate = (filp->f_version != inode->i_version);
310 CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p) pos %llu/%llu\n",
311 inode->i_ino, inode->i_generation, inode, pos, inode->i_size);
313 if (pos > inode->i_size - EXT2_DIR_REC_LEN(1))
316 types = ext2_filetype_table;
318 for ( ; n < npages; n++, offset = 0) {
323 CDEBUG(D_EXT2,"read %lu of dir %lu/%u page %lu/%lu size %llu\n",
324 PAGE_CACHE_SIZE, inode->i_ino, inode->i_generation,
325 n, npages, inode->i_size);
326 page = ll_get_dir_page(inode, n);
328 /* size might have been updated by md_readpage */
329 npages = dir_pages(inode);
333 CERROR("error reading dir %lu/%u page %lu: rc %d\n",
334 inode->i_ino, inode->i_generation, n, rc);
338 kaddr = page_address(page);
339 if (need_revalidate) {
340 offset = ext2_validate_entry(kaddr, offset, chunk_mask);
343 de = (ext2_dirent *)(kaddr+offset);
344 limit = kaddr + PAGE_CACHE_SIZE - EXT2_DIR_REC_LEN(1);
345 for ( ;(char*)de <= limit; de = ext2_next_entry(de)) {
348 unsigned char d_type = DT_UNKNOWN;
350 rc = 0; /* no error if we return something */
351 if (types && de->file_type < EXT2_FT_MAX)
352 d_type = types[de->file_type];
354 offset = (char *)de - kaddr;
355 over = filldir(dirent, de->name, de->name_len,
356 (n<<PAGE_CACHE_SHIFT) | offset,
357 le32_to_cpu(de->inode), d_type);
368 filp->f_pos = (n << PAGE_CACHE_SHIFT) | offset;
369 filp->f_version = inode->i_version;
374 static int ll_mkdir_stripe(struct inode *inode, unsigned long arg)
377 struct ll_user_mkdir_stripe lums;
378 struct ptlrpc_request *request = NULL;
379 struct ll_sb_info *sbi = ll_i2sbi(inode);
380 struct mdc_op_data op_data;
386 if (copy_from_user(&lums, (void *)arg, sizeof(lums)))
389 if (lums.lums_namelen <= 0)
391 OBD_ALLOC(name, lums.lums_namelen);
395 if (copy_from_user(name, lums.lums_name, lums.lums_namelen))
396 GOTO(out, err=-EFAULT);
398 CDEBUG(D_VFSTRACE, "ioctl Op:name=%s,dir=%lu/%u(%p)\n",
399 name, inode->i_ino, inode->i_generation, inode);
400 nstripes = lums.lums_nstripes;
402 mode = lums.lums_mode;
403 mode = (mode & (S_IRWXUGO|S_ISVTX) & ~current->fs->umask) | S_IFDIR;
404 ll_prepare_mdc_op_data(&op_data, inode, NULL, name,lums.lums_namelen,0);
405 err = md_create(sbi->ll_mdc_exp, &op_data, &nstripes, sizeof(nstripes),
406 mode, current->fsuid, current->fsgid, 0, &request);
407 ptlrpc_req_finished(request);
410 OBD_FREE(name, lums.lums_namelen);
414 static int ll_dir_ioctl(struct inode *inode, struct file *file,
415 unsigned int cmd, unsigned long arg)
417 struct ll_sb_info *sbi = ll_i2sbi(inode);
418 struct obd_ioctl_data *data;
421 CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p), cmd=%#x\n",
422 inode->i_ino, inode->i_generation, inode, cmd);
424 if (_IOC_TYPE(cmd) == 'T') /* tty ioctls */
427 lprocfs_counter_incr(ll_i2sbi(inode)->ll_stats, LPROC_LL_IOCTL);
429 case EXT3_IOC_GETFLAGS:
430 case EXT3_IOC_SETFLAGS:
431 RETURN( ll_iocontrol(inode, file, cmd, arg) );
432 case IOC_MDC_LOOKUP: {
433 struct ptlrpc_request *request = NULL;
437 int namelen, rc, len = 0;
440 rc = obd_ioctl_getdata(&buf, &len, (void *)arg);
445 filename = data->ioc_inlbuf1;
446 namelen = data->ioc_inllen1;
449 CDEBUG(D_INFO, "IOC_MDC_LOOKUP missing filename\n");
450 GOTO(out, rc = -EINVAL);
454 ll_inode2fid(&fid, inode);
455 rc = md_getattr_name(sbi->ll_mdc_exp, &fid,
456 filename, namelen, valid,
459 CDEBUG(D_INFO, "md_getattr_name: %d\n", rc);
463 ptlrpc_req_finished(request);
467 obd_ioctl_freedata(buf, len);
470 case LL_IOC_MDC_MKDIRSTRIPE:
471 RETURN(ll_mkdir_stripe(inode, arg));
472 case LL_IOC_LOV_SETSTRIPE:
473 case LL_IOC_LOV_GETSTRIPE:
475 case IOC_MDC_GETSTRIPE: {
476 struct ptlrpc_request *request = NULL;
478 struct mds_body *body;
479 struct lov_user_md *lump = (struct lov_user_md *)arg;
480 struct lov_mds_md *lmm;
484 filename = getname((const char *)arg);
485 if (IS_ERR(filename))
486 RETURN(PTR_ERR(filename));
488 ll_inode2fid(&fid, inode);
489 rc = md_getattr_name(sbi->ll_mdc_exp, &fid, filename,
490 strlen(filename)+1, OBD_MD_FLEASIZE,
491 obd_size_diskmd(sbi->ll_osc_exp, NULL),
495 "md_getattr_name failed on %s: rc %d\n",
500 body = lustre_msg_buf(request->rq_repmsg, 0, sizeof (*body));
501 LASSERT(body != NULL); /* checked by md_getattr_name */
502 LASSERT_REPSWABBED(request, 0);/* swabbed by mdc_getattr_name */
504 lmmsize = body->eadatasize;
506 if (!(body->valid & OBD_MD_FLEASIZE) || lmmsize == 0)
507 GOTO(out_req, rc = -ENODATA);
510 GOTO(out_req, rc = -EFBIG);
512 lmm = lustre_msg_buf(request->rq_repmsg, 1, lmmsize);
513 LASSERT(lmm != NULL);
514 LASSERT_REPSWABBED(request, 1);
516 rc = copy_to_user(lump, lmm, lmmsize);
518 GOTO(out_req, rc = -EFAULT);
522 ptlrpc_req_finished(request);
528 struct ptlrpc_request *req = NULL;
531 struct client_obd *cli;
532 struct obd_device *obd;
534 rc = obd_ioctl_getdata(&buf, &len, (void *)arg);
539 obd = class_name2obd(data->ioc_inlbuf1);
542 GOTO(out_ping, rc = -ENODEV);
544 if (!obd->obd_attached) {
545 CERROR("Device %d not attached\n", obd->obd_minor);
546 GOTO(out_ping, rc = -ENODEV);
548 if (!obd->obd_set_up) {
549 CERROR("Device %d still not setup\n", obd->obd_minor);
550 GOTO(out_ping, rc = -ENODEV);
553 req = ptlrpc_prep_req(cli->cl_import, OBD_PING, 0, NULL, NULL);
555 GOTO(out_ping, rc = -ENOMEM);
557 req->rq_replen = lustre_msg_size(0, NULL);
558 req->rq_send_state = LUSTRE_IMP_FULL;
560 rc = ptlrpc_queue_wait(req);
562 ptlrpc_req_finished(req);
564 obd_ioctl_freedata(buf, len);
567 case OBD_IOC_LLOG_CATINFO: {
568 struct ptlrpc_request *req = NULL;
574 rc = obd_ioctl_getdata(&buf, &len, (void *)arg);
579 if (!data->ioc_inlbuf1) {
580 obd_ioctl_freedata(buf, len);
584 lens[0] = data->ioc_inllen1;
585 bufs[0] = data->ioc_inlbuf1;
586 if (data->ioc_inllen2) {
587 lens[1] = data->ioc_inllen2;
588 bufs[1] = data->ioc_inlbuf2;
593 size = data->ioc_plen1;
594 req = ptlrpc_prep_req(sbi2mdc(sbi)->cl_import, LLOG_CATINFO,
597 GOTO(out_catinfo, rc = -ENOMEM);
598 req->rq_replen = lustre_msg_size(1, &size);
600 rc = ptlrpc_queue_wait(req);
601 str = lustre_msg_string(req->rq_repmsg, 0, data->ioc_plen1);
603 rc = copy_to_user(data->ioc_pbuf1, str,
605 ptlrpc_req_finished(req);
607 obd_ioctl_freedata(buf, len);
611 return obd_iocontrol(cmd, sbi->ll_osc_exp,0,NULL,(void *)arg);
615 int ll_dir_open(struct inode *inode, struct file *file)
617 return ll_file_open(inode, file);
620 int ll_dir_release(struct inode *inode, struct file *file)
622 return ll_file_release(inode, file);
625 struct file_operations ll_dir_operations = {
627 release: ll_dir_release,
628 read: generic_read_dir,