log write() data

usertest for big write()s push begin_trans/commit_trans down into syscalls

log write() data
2e590463 · Robert Morris · bd71a450 · 2e590463 · 2e590463 · 2e590463
--- a/file.c
+++ b/file.c
@@ -67,8 +67,11 @@ fileclose(struct file *f)
  if(ff.type == FD_PIPE)
    pipeclose(ff.pipe, ff.writable);
-  else if(ff.type == FD_INODE)
+  else if(ff.type == FD_INODE){
+    begin_trans();
    iput(ff.ip);
+    commit_trans();
+  }
 }
 // Get metadata about file f.
@@ -116,10 +119,30 @@ filewrite(struct file *f, char *addr, int n)
    return pipewrite(f->pipe, addr, n);
  if(f->type == FD_INODE){
    ilock(f->ip);
-    if((r = writei(f->ip, addr, f->off, n)) > 0)
+    // write a few blocks at a time to avoid exceeding
+    // the maximum log transaction size, including
+    // i-node, indirect block, allocation blocks,
+    // and 2 blocks of slop for non-aligned writes.
+    // this really belongs lower down, since writei()
+    // might be writing a device like the console.
+    int max = ((LOGSIZE-1-1-2) / 2) * 512;
+    int i = 0;
+    while(i < n){
+      int n1 = n - i;
+      if(n1 > max)
+        n1 = max;
+      begin_trans();
+      r = writei(f->ip, addr + i, f->off, n1);
+      commit_trans();
+      if(r < 0)
+        break;
+      if(r != n1)
+        panic("short filewrite");
      f->off += r;
+      i += r;
+    }
    iunlock(f->ip);
-    return r;
+    return i == n ? n : -1;
  }
  panic("filewrite");
 }

--- a/fs.c
+++ b/fs.c
@@ -437,13 +437,13 @@ writei(struct inode *ip, char *src, uint off, uint n)
  if(off > ip->size || off + n < off)
    return -1;
  if(off + n > MAXFILE*BSIZE)
-    n = MAXFILE*BSIZE - off;
+    return -1;
  for(tot=0; tot<n; tot+=m, off+=m, src+=m){
    bp = bread(ip->dev, bmap(ip, off/BSIZE));
    m = min(n - tot, BSIZE - off%BSIZE);
    memmove(bp->data + off%BSIZE, src, m);
-    bwrite(bp);
+    log_write(bp);
    brelse(bp);
  }

--- a/log.c
+++ b/log.c
@@ -8,18 +8,36 @@
 #include "fs.h"
 #include "buf.h"
-// Dirt simple "logging" supporting only one transaction.  All file system calls
+// Simple logging. Each system call that might write the file system
-// that potentially write a block should be wrapped in begin_trans and commit_trans,
+// should be surrounded with begin_trans() and commit_trans() calls.
-// so that there is never more than one transaction. This serializes all file system 
+//
-// operations that potentially write, but simplifies recovery (only the last
+// The log holds at most one transaction at a time. Commit forces
-// one transaction to recover) and concurrency (don't have to worry about reading a modified
+// the log (with commit record) to disk, then installs the affected
-// block from a transaction that hasn't committed yet).
+// blocks to disk, then erases the log. begin_trans() ensures that
+// only one system call can be in a transaction; others must wait.
-// The header of the log.  If head == 0, there are no log entries.  All entries till head
+// 
-// are committed. sector[] records the home sector for each block in the log 
+// Allowing only one transaction at a time means that the file
-// (i.e., physical logging).
+// system code doesn't have to worry about the possibility of
+// one transaction reading a block that another one has modified,
+// for example an i-node block.
+//
+// Read-only system calls don't need to use transactions, though
+// this means that they may observe uncommitted data. I-node
+// and buffer locks prevent read-only calls from seeing inconsistent data.
+//
+// The log is a physical re-do log containing disk blocks.
+// The on-disk log format:
+//   header block, containing sector #s for block A, B, C, ...
+//   block A
+//   block B
+//   block C
+//   ...
+// Log appends are synchronous.
+// Contents of the header block, used for both the on-disk header block
+// and to keep track in memory of logged sector #s before commit.
 struct logheader {
-  int head;   
+  int n;   
  int sector[LOGSIZE];
 };
@@ -55,10 +73,10 @@ install_trans(void)
 {
  int tail;
-  if (log.lh.head > 0)
+  //if (log.lh.n > 0)
-    cprintf("install_trans %d\n", log.lh.head);
+  //  cprintf("install_trans %d\n", log.lh.n);
-  for (tail = 0; tail < log.lh.head; tail++) {
+  for (tail = 0; tail < log.lh.n; tail++) {
-    cprintf("put entry %d to disk block %d\n", tail, log.lh.sector[tail]);
+    // cprintf("put entry %d to disk block %d\n", tail, log.lh.sector[tail]);
    struct buf *lbuf = bread(log.dev, log.start+tail+1);   // read i'th block from log
    struct buf *dbuf = bread(log.dev, log.lh.sector[tail]);  // read dst block
    memmove(dbuf->data, lbuf->data, BSIZE);
@@ -75,27 +93,27 @@ read_head(void)
  struct buf *buf = bread(log.dev, log.start);
  struct logheader *lh = (struct logheader *) (buf->data);
  int i;
-  log.lh.head = lh->head;
+  log.lh.n = lh->n;
-  for (i = 0; i < log.lh.head; i++) {
+  for (i = 0; i < log.lh.n; i++) {
    log.lh.sector[i] = lh->sector[i];
  }
  brelse(buf);
-  if (log.lh.head > 0)
+  //if (log.lh.n > 0)
-    cprintf("read_head: %d\n", log.lh.head);
+  //  cprintf("read_head: %d\n", log.lh.n);
 }
 // Write the in-memory log header to disk, committing log entries till head
 static void
 write_head(void)
 {
-  if (log.lh.head > 0)
+  // if (log.lh.n > 0)
-    cprintf("write_head: %d\n", log.lh.head);
+  //   cprintf("write_head: %d\n", log.lh.n);
  struct buf *buf = bread(log.dev, log.start);
  struct logheader *hb = (struct logheader *) (buf->data);
  int i;
-  hb->head = log.lh.head;
+  hb->n = log.lh.n;
-  for (i = 0; i < log.lh.head; i++) {
+  for (i = 0; i < log.lh.n; i++) {
    hb->sector[i] = log.lh.sector[i];
  }
  bwrite(buf);
@@ -107,7 +125,7 @@ recover_from_log(void)
 {
  read_head();      
  install_trans();  // Install all transactions till head
-  log.lh.head = 0;
+  log.lh.n = 0;
  write_head();     //  Reclaim log
 }
@@ -127,7 +145,7 @@ commit_trans(void)
 {
  write_head();        // This causes all blocks till log.head to be commited
  install_trans();     // Install all the transactions till head
-  log.lh.head = 0; 
+  log.lh.n = 0; 
  write_head();        // Reclaim log
  acquire(&log.lock);
@@ -136,21 +154,27 @@ commit_trans(void)
  release(&log.lock);
 }
-// Write buffer into the log at log.head and record the block number log.lh.entry, but
+// Caller has modified b->data and is done with the buffer.
-// don't write the log header (which would commit the write).
+// Append the block to the log and record the block number, 
+// but don't write the log header (which would commit the write).
+// log_write() replaces bwrite(); a typical use is:
+//   bp = bread(...)
+//   modify bp->data[]
+//   log_write(bp)
+//   brelse(bp)
 void
 log_write(struct buf *b)
 {
  int i;
-  if (log.lh.head >= LOGSIZE)
+  if (log.lh.n >= LOGSIZE || log.lh.n >= log.size - 1)
    panic("too big a transaction");
  if (!log.intrans)
    panic("write outside of trans");
-  cprintf("log_write: %d %d\n", b->sector, log.lh.head);
+  // cprintf("log_write: %d %d\n", b->sector, log.lh.n);
-  for (i = 0; i < log.lh.head; i++) {
+  for (i = 0; i < log.lh.n; i++) {
    if (log.lh.sector[i] == b->sector)   // log absorbtion?
      break;
  }
@@ -159,6 +183,6 @@ log_write(struct buf *b)
  memmove(lbuf->data, b->data, BSIZE);
  bwrite(lbuf);
  brelse(lbuf);
-  if (i == log.lh.head)
+  if (i == log.lh.n)
-    log.lh.head++;
+    log.lh.n++;
 }
--- a/param.h
+++ b/param.h
@@ -8,5 +8,5 @@
 #define NDEV         10  // maximum major device number
 #define ROOTDEV       1  // device number of file system root disk
 #define MAXARG       32  // max exec arguments
-#define LOGSIZE      10  // size of log
+#define LOGSIZE      10  // max data sectors in on-disk log
--- a/syscall.c
+++ b/syscall.c
@@ -141,9 +141,7 @@ syscall(void)
  if(num >= 0 && num < SYS_open && syscalls[num]) {
    proc->tf->eax = syscalls[num]();
  } else if (num >= SYS_open && num < NELEM(syscalls) && syscalls[num]) {
-    begin_trans();
    proc->tf->eax = syscalls[num]();
-    commit_trans();
  } else {
    cprintf("%d %s: unknown sys call %d\n",
            proc->pid, proc->name, num);

--- a/sysfile.c
+++ b/sysfile.c
@@ -121,6 +121,9 @@ sys_link(void)
    iunlockput(ip);
    return -1;
  }
+  begin_trans();
  ip->nlink++;
  iupdate(ip);
  iunlock(ip);
@@ -134,6 +137,9 @@ sys_link(void)
  }
  iunlockput(dp);
  iput(ip);
+  commit_trans();
  return 0;
 bad:
@@ -141,6 +147,7 @@ bad:
  ip->nlink--;
  iupdate(ip);
  iunlockput(ip);
+  commit_trans();
  return -1;
 }
@@ -195,6 +202,8 @@ sys_unlink(void)
    return -1;
  }
+  begin_trans();
  memset(&de, 0, sizeof(de));
  if(writei(dp, (char*)&de, off, sizeof(de)) != sizeof(de))
    panic("unlink: writei");
@@ -207,6 +216,9 @@ sys_unlink(void)
  ip->nlink--;
  iupdate(ip);
  iunlockput(ip);
+  commit_trans();
  return 0;
 }
@@ -251,6 +263,7 @@ create(char *path, short type, short major, short minor)
    panic("create: dirlink");
  iunlockput(dp);
  return ip;
 }
@@ -265,7 +278,10 @@ sys_open(void)
  if(argstr(0, &path) < 0 || argint(1, &omode) < 0)
    return -1;
  if(omode & O_CREATE){
-    if((ip = create(path, T_FILE, 0, 0)) == 0)
+    begin_trans();
+    ip = create(path, T_FILE, 0, 0);
+    commit_trans();
+    if(ip == 0)
      return -1;
  } else {
    if((ip = namei(path)) == 0)
@@ -299,9 +315,13 @@ sys_mkdir(void)
  char *path;
  struct inode *ip;
-  if(argstr(0, &path) < 0 || (ip = create(path, T_DIR, 0, 0)) == 0)
+  begin_trans();
+  if(argstr(0, &path) < 0 || (ip = create(path, T_DIR, 0, 0)) == 0){
+    commit_trans();
    return -1;
+  }
  iunlockput(ip);
+  commit_trans();
  return 0;
 }
@@ -313,12 +333,16 @@ sys_mknod(void)
  int len;
  int major, minor;
+  begin_trans();
  if((len=argstr(0, &path)) < 0 ||
     argint(1, &major) < 0 ||
     argint(2, &minor) < 0 ||
-     (ip = create(path, T_DEV, major, minor)) == 0)
+     (ip = create(path, T_DEV, major, minor)) == 0){
+    commit_trans();
    return -1;
+  }
  iunlockput(ip);
+  commit_trans();
  return 0;
 }

--- a/usertests.c
+++ b/usertests.c
@@ -7,7 +7,7 @@
 #include "traps.h"
 #include "memlayout.h"
-char buf[2048];
+char buf[8192];
 char name[3];
 char *echoargv[] = { "echo", "ALL", "TESTS", "PASSED", 0 };
 int stdout = 1;
@@ -968,6 +968,36 @@ subdir(void)
  printf(1, "subdir ok\n");
 }
+// test writes that are larger than the log.
+void
+bigwrite(void)
+{
+  int fd, sz;
+  printf(1, "bigwrite test\n");
+  unlink("bigwrite");
+  for(sz = 499; sz < 12*512; sz += 471){
+    fd = open("bigwrite", O_CREATE | O_RDWR);
+    if(fd < 0){
+      printf(1, "cannot create bigwrite\n");
+      exit();
+    }
+    int i;
+    for(i = 0; i < 2; i++){
+      int cc = write(fd, buf, sz);
+      if(cc != sz){
+        printf(1, "write(%d) ret %d\n", sz, cc);
+        exit();
+      }
+    }
+    close(fd);
+    unlink("bigwrite");
+  }
+  printf(1, "bigwrite ok\n");
+}
 void
 bigfile(void)
 {
@@ -1467,6 +1497,7 @@ main(int argc, char *argv[])
  }
  close(open("usertests.ran", O_CREATE));
+  bigwrite();
  bigargtest();
  bsstest();
  sbrktest();