lustre/llite/rw.c

   1 /* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
   2  * vim:expandtab:shiftwidth=8:tabstop=8:
   3  *
   4  * Lustre Lite I/O Page Cache
   5  *
   6  * Copyright (C) 2002 Cluster File Systems, Inc.
   7  */
   8
   9 #include <linux/config.h>
  10 #include <linux/kernel.h>
  11 #include <linux/mm.h>
  12 #include <linux/string.h>
  13 #include <linux/stat.h>
  14 #include <linux/iobuf.h>
  15 #include <linux/errno.h>
  16 #include <linux/smp_lock.h>
  17 #include <linux/unistd.h>
  18 #include <linux/version.h>
  19 #include <asm/system.h>
  20 #include <asm/uaccess.h>
  21
  22 #include <linux/fs.h>
  23 #include <linux/stat.h>
  24 #include <asm/uaccess.h>
  25 #include <asm/segment.h>
  26 #include <linux/mm.h>
  27 #include <linux/pagemap.h>
  28 #include <linux/smp_lock.h>
  29
  30 #define DEBUG_SUBSYSTEM S_LLITE
  31
  32 #include <linux/lustre_mds.h>
  33 #include <linux/lustre_lite.h>
  34 #include <linux/lustre_lib.h>
  35
  36 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,5,0))
  37
  38 /*
  39  * Remove page from dirty list
  40  */
  41 static void __set_page_clean(struct page *page)
  42 {
  43         struct address_space *mapping = page->mapping;
  44         struct inode *inode;
  45
  46         if (!mapping)
  47                 return;
  48
  49 #if (LINUX_VERSION_CODE > KERNEL_VERSION(2,4,9))
  50         spin_lock(&pagecache_lock);
  51 #endif
  52
  53         list_del(&page->list);
  54         list_add(&page->list, &mapping->clean_pages);
  55
  56         inode = mapping->host;
  57         if (list_empty(&mapping->dirty_pages)) {
  58                 CDEBUG(D_INODE, "inode clean\n");
  59                 inode->i_state &= ~I_DIRTY_PAGES;
  60         }
  61 #if (LINUX_VERSION_CODE > KERNEL_VERSION(2,4,10))
  62         spin_unlock(&pagecache_lock);
  63 #endif
  64         EXIT;
  65 }
  66
  67 inline void set_page_clean(struct page *page)
  68 {
  69         if (PageDirty(page)) {
  70                 ClearPageDirty(page);
  71                 __set_page_clean(page);
  72         }
  73 }
  74
  75 /* SYNCHRONOUS I/O to object storage for an inode */
  76 static int ll_brw(int cmd, struct inode *inode, struct page *page, int create)
  77 {
  78         struct ll_inode_info *lli = ll_i2info(inode);
  79         struct lov_stripe_md *lsm = lli->lli_smd;
  80         struct io_cb_data *cbd = ll_init_cb();
  81         struct brw_page pg;
  82         int err;
  83         ENTRY;
  84
  85         CHECK_MOUNT_EPOCH(inode);
  86
  87         if (!cbd)
  88                 RETURN(-ENOMEM);
  89
  90         pg.pg = page;
  91         pg.count = PAGE_SIZE;
  92         pg.off = ((obd_off)page->index) << PAGE_SHIFT;
  93         pg.flag = create ? OBD_BRW_CREATE : 0;
  94
  95         err = obd_brw(cmd, ll_i2obdconn(inode),lsm, 1, &pg, ll_sync_io_cb, cbd);
  96
  97         RETURN(err);
  98 } /* ll_brw */
  99
 100 /* returns the page unlocked, but with a reference */
 101 static int ll_readpage(struct file *file, struct page *page)
 102 {
 103         struct inode *inode = page->mapping->host;
 104         obd_off offset = ((obd_off)page->index) << PAGE_SHIFT;
 105         int rc = 0;
 106         ENTRY;
 107
 108         if (!PageLocked(page))
 109                 LBUG();
 110
 111         if (inode->i_size <= offset) {
 112                 memset(kmap(page), 0, PAGE_SIZE);
 113                 kunmap(page);
 114                 GOTO(readpage_out, rc);
 115         }
 116
 117         if (Page_Uptodate(page)) {
 118                 CERROR("Explain this please?\n");
 119                 GOTO(readpage_out, rc);
 120         }
 121
 122         rc = ll_brw(OBD_BRW_READ, inode, page, 0);
 123         EXIT;
 124
 125  readpage_out:
 126         if (!rc)
 127                 SetPageUptodate(page);
 128         unlock_page(page);
 129         return 0;
 130 } /* ll_readpage */
 131
 132
 133 static int ll_prepare_write(struct file *file, struct page *page, unsigned from,
 134                             unsigned to)
 135 {
 136         struct inode *inode = page->mapping->host;
 137         obd_off offset = ((obd_off)page->index) << PAGE_SHIFT;
 138         int rc = 0;
 139         char *addr;
 140         ENTRY;
 141
 142         addr = kmap(page);
 143         if (!PageLocked(page))
 144                 LBUG();
 145
 146         if (Page_Uptodate(page))
 147                 GOTO(prepare_done, rc);
 148
 149         /* We're completely overwriting an existing page, so _don't_ set it up
 150          * to date until commit_write */
 151         if (from == 0 && to == PAGE_SIZE)
 152                 RETURN(0);
 153
 154         /* We are writing to a new page, no need to read old data */
 155         if (inode->i_size <= offset) {
 156                 memset(addr, 0, PAGE_SIZE);
 157                 goto prepare_done;
 158         }
 159
 160         rc = ll_brw(OBD_BRW_READ, inode, page, 0);
 161
 162         EXIT;
 163  prepare_done:
 164         if (!rc)
 165                 SetPageUptodate(page);
 166
 167         return rc;
 168 }
 169
 170 /* returns the page unlocked, but with a reference */
 171 static int ll_writepage(struct page *page)
 172 {
 173         struct inode *inode = page->mapping->host;
 174         int err;
 175         ENTRY;
 176
 177         if (!PageLocked(page))
 178                 LBUG();
 179
 180         err = ll_brw(OBD_BRW_WRITE, inode, page, 1);
 181         if ( !err ) {
 182                 //SetPageUptodate(page);
 183                 set_page_clean(page);
 184         } else {
 185                 CERROR("ll_brw failure %d\n", err);
 186         }
 187         unlock_page(page);
 188         RETURN(err);
 189 }
 190
 191
 192 /* SYNCHRONOUS I/O to object storage for an inode -- object attr will be updated
 193  * too */
 194 static int ll_commit_write(struct file *file, struct page *page,
 195                            unsigned from, unsigned to)
 196 {
 197         int create = 1;
 198         struct inode *inode = page->mapping->host;
 199         struct ll_inode_info *lli = ll_i2info(inode);
 200         struct lov_stripe_md *md = lli->lli_smd;
 201         struct brw_page pg;
 202         int err;
 203         loff_t size;
 204         struct io_cb_data *cbd = ll_init_cb();
 205         ENTRY;
 206
 207         CHECK_MOUNT_EPOCH(inode);
 208
 209         pg.pg = page;
 210         pg.count = to;
 211         pg.off = (((obd_off)page->index) << PAGE_SHIFT);
 212         pg.flag = create ? OBD_BRW_CREATE : 0;
 213
 214         if (!cbd)
 215                 RETURN(-ENOMEM);
 216
 217         SetPageUptodate(page);
 218
 219         if (!PageLocked(page))
 220                 LBUG();
 221
 222         CDEBUG(D_INODE, "commit_page writing (off "LPD64"), count "LPD64"\n",
 223                pg.off, pg.count);
 224
 225         err = obd_brw(OBD_BRW_WRITE, ll_i2obdconn(inode), md,
 226                       1, &pg, ll_sync_io_cb, cbd);
 227         kunmap(page);
 228
 229         size = pg.off + pg.count;
 230         /* do NOT truncate when writing in the middle of a file */
 231         if (size > inode->i_size)
 232                 inode->i_size = size;
 233
 234         RETURN(err);
 235 } /* ll_commit_write */
 236
 237 void ll_truncate(struct inode *inode)
 238 {
 239         struct obdo oa = {0};
 240         struct lov_stripe_md *lsm = ll_i2info(inode)->lli_smd;
 241         struct lustre_handle *lockhs = NULL;
 242         int err;
 243         ENTRY;
 244
 245         if (!lsm) {
 246                 /* object not yet allocated */
 247                 inode->i_mtime = inode->i_ctime = CURRENT_TIME;
 248                 return;
 249         }
 250
 251         oa.o_id = lsm->lsm_object_id;
 252         oa.o_mode = inode->i_mode;
 253         oa.o_valid = OBD_MD_FLID | OBD_MD_FLMODE | OBD_MD_FLTYPE;
 254
 255         CDEBUG(D_INFO, "calling punch for "LPX64" (all bytes after "LPD64")\n",
 256                oa.o_id, inode->i_size);
 257
 258         err = ll_size_lock(inode, lsm, inode->i_size, LCK_PW, &lockhs);
 259         if (err) {
 260                 CERROR("ll_size_lock failed: %d\n", err);
 261                 /* FIXME: What to do here?  It's too late to back out... */
 262                 LBUG();
 263         }
 264
 265         /* truncate == punch from new size to absolute end of file */
 266         err = obd_punch(ll_i2obdconn(inode), &oa, lsm, inode->i_size,
 267                         OBD_OBJECT_EOF);
 268         if (err)
 269                 CERROR("obd_truncate fails (%d)\n", err);
 270         else
 271                 obdo_to_inode(inode, &oa, oa.o_valid);
 272
 273         err = ll_size_unlock(inode, lsm, LCK_PW, lockhs);
 274         if (err)
 275                 CERROR("ll_size_unlock failed: %d\n", err);
 276
 277         EXIT;
 278         return;
 279 } /* ll_truncate */
 280
 281 static int ll_direct_IO(int rw, struct inode *inode, struct kiobuf *iobuf,
 282                         unsigned long blocknr, int blocksize)
 283 {
 284         obd_count bufs_per_obdo = iobuf->nr_pages;
 285         struct ll_inode_info *lli = ll_i2info(inode);
 286         struct lov_stripe_md *lsm = lli->lli_smd;
 287         struct brw_page *pga;
 288         int i, rc = 0;
 289         struct io_cb_data *cbd;
 290
 291         CHECK_MOUNT_EPOCH(inode);
 292
 293         ENTRY;
 294         if (!lsm || !lsm->lsm_object_id)
 295                 RETURN(-ENOMEM);
 296
 297         if (blocksize != PAGE_SIZE) {
 298                 CERROR("direct_IO blocksize != PAGE_SIZE\n");
 299                 RETURN(-EINVAL);
 300         }
 301
 302         cbd = ll_init_cb();
 303         if (!cbd)
 304                 RETURN(-ENOMEM);
 305
 306         OBD_ALLOC(pga, sizeof(*pga) * bufs_per_obdo);
 307         if (!pga) {
 308                 OBD_FREE(cbd, sizeof(*cbd));
 309                 RETURN(-ENOMEM);
 310         }
 311
 312         /* NB: we can't use iobuf->maplist[i]->index for the offset
 313          * instead of "blocknr" because ->index contains garbage.
 314          */
 315         for (i = 0; i < bufs_per_obdo; i++, blocknr++) {
 316                 pga[i].pg = iobuf->maplist[i];
 317                 pga[i].count = PAGE_SIZE;
 318                 pga[i].off = (obd_off)blocknr << PAGE_SHIFT;
 319                 pga[i].flag = OBD_BRW_CREATE;
 320         }
 321
 322         rc = obd_brw(rw == WRITE ? OBD_BRW_WRITE : OBD_BRW_READ,
 323                      ll_i2obdconn(inode), lsm, bufs_per_obdo, pga,
 324                      ll_sync_io_cb, cbd);
 325         if (rc == 0)
 326                 rc = bufs_per_obdo * PAGE_SIZE;
 327
 328         OBD_FREE(pga, sizeof(*pga) * bufs_per_obdo);
 329         RETURN(rc);
 330 }
 331
 332 int ll_flush_inode_pages(struct inode * inode)
 333 {
 334         obd_count        bufs_per_obdo = 0;
 335         obd_size         *count = NULL;
 336         obd_off          *offset = NULL;
 337         obd_flag         *flags = NULL;
 338         int              err = 0;
 339
 340         ENTRY;
 341
 342         spin_lock(&pagecache_lock);
 343
 344         spin_unlock(&pagecache_lock);
 345
 346
 347         OBD_ALLOC(count, sizeof(*count) * bufs_per_obdo);
 348         OBD_ALLOC(offset, sizeof(*offset) * bufs_per_obdo);
 349         OBD_ALLOC(flags, sizeof(*flags) * bufs_per_obdo);
 350         if (!count || !offset || !flags)
 351                 GOTO(out, err=-ENOMEM);
 352
 353 #if 0
 354         for (i = 0 ; i < bufs_per_obdo ; i++) {
 355                 count[i] = PAGE_SIZE;
 356                 offset[i] = ((obd_off)(iobuf->maplist[i])->index) << PAGE_SHIFT;
 357                 flags[i] = OBD_BRW_CREATE;
 358         }
 359
 360         err = obd_brw(OBD_BRW_WRITE, ll_i2obdconn(inode),
 361                       ll_i2info(inode)->lli_smd, bufs_per_obdo,
 362                       iobuf->maplist, count, offset, flags, NULL, NULL);
 363         if (err == 0)
 364                 err = bufs_per_obdo * 4096;
 365 #endif
 366  out:
 367         OBD_FREE(flags, sizeof(*flags) * bufs_per_obdo);
 368         OBD_FREE(count, sizeof(*count) * bufs_per_obdo);
 369         OBD_FREE(offset, sizeof(*offset) * bufs_per_obdo);
 370         RETURN(err);
 371 }
 372
 373 struct address_space_operations ll_aops = {
 374         readpage: ll_readpage,
 375         writepage: ll_writepage,
 376 #if (LINUX_VERSION_CODE > KERNEL_VERSION(2,4,17))
 377         direct_IO: ll_direct_IO,
 378 #endif
 379         sync_page: block_sync_page,
 380         prepare_write: ll_prepare_write,
 381         commit_write: ll_commit_write,
 382         bmap: NULL
 383 };
 384 #endif